AI does the heavy lifting with content and site checks, but human expertise makes the winning decisions. It takes real human judgment to know what strategy to use, when to use it, and why.
24 MIN READ14 VISUAL EXPLAINERS6 INTERACTIVE LABSREAL CLAUDE EXPERIMENT
A conversation is becoming more common between business owners and SEO agencies. It usually starts with a sentence that sounds completely reasonable:
“I think I’ll keep the backlinks package. I can use AI for everything else.”
And there is a good reason that idea is appealing. ChatGPT can write content, Claude can audit pages, Gemini can research, coding agents can generate schema or technical fixes, and modern AI systems can process in minutes what used to take a marketing team hours.
So instead of arguing that AI cannot do SEO, we wanted to test the more useful question: if AI can now do so much of the execution, what exactly is left for the consultant to do?
“
If AI could do it all, everyone using AI would rank in the first position. But your competitors have access to the same AI and the same tools. The real advantage is knowing how to use them better — what to prioritize, what to ignore and where the opportunity actually is. This is where strategy comes in, and that strategy comes from years of experience and expertise running marketing campaigns across markets worldwide.
Ishan Gupta, CEO, iMark Infotech
The answer cannot simply be “experience” or “strategy” because those words are too easy to say. If a consultant still matters, we should be able to show where that value appears in the work itself. That is why we started with an experiment rather than a sales argument.
Before the Experiment, We Had Already Seen the Question in the Real World
One of our clients had already tried taking more of the SEO process into his own hands with AI. The tools were useful, and that part matters: this was not a story about AI being incapable. What became difficult was deciding which recommendations deserved action, how technical changes should actually be implemented, and whether the work was improving the business rather than simply producing more SEO activity.
Client perspective
Recommendations are easy. Knowing what to do next is harder.
Melton shares what happened when he tried to take more of the SEO process into his own hands with AI.
Temporary video preview
That experience gave us a simple hypothesis to test: AI may be very good at producing SEO work, while still leaving the harder decision of what deserves to be done unresolved.
We Gave Claude the Job — Then Asked Claude to Audit It
We used our own iMark Infotech homepage content so we would not be commenting on somebody else’s campaign. In the first fresh conversation, we asked Claude to “Optimize this page's content perfectly for SEO.” It selected a direction, recommended changes to the title, H1, homepage positioning, internal linking, structured data, FAQs and content structure, and then rewrote the page. You can inspect Claude’s original optimization conversation.
Then we opened a completely fresh chat, gave Claude the content it had just created and asked it to audit the page as critically as possible. This time it found problems with the earlier work, questioned the title and H1 strategy, and described one decision as a “self-inflicted wound.” We let it revise the page again. The full second Claude audit is public here.
Then we repeated the process a third time in another clean conversation. One of the first questions Claude raised was whether the “primary keyword may not be worth optimizing for.” You can read the third fresh audit here.
Claude chose the direction, criticized the execution, fixed the execution, and then questioned the direction itself.
That does not prove Claude is bad at SEO. In fact, the opposite is closer to what we observed: it found useful issues every time. The interesting part was that there was no natural stopping point. Every fresh instruction to “find more problems” produced more work to consider.
Eventually somebody has to say: this is good enough; the next hour is more valuable somewhere else. Optimization without prioritization can become its own form of waste.
Research system / AI recommendation consistencySystem ready
iMark Infotech · Original experiment · 2026
We asked Claude to audit Claude.
Three isolated conversations. One homepage. Each fresh context found a new reason to change what the previous model had produced.
OUTPUTKeyword direction selected. “Optimized” homepage copy generated.
02
Context reset / complete
Run 02 · AuditT+02:40
Challenge
“Title/H1 mismatch. Splitting keyword equity is a self-inflicted wound.”
NEW FINDINGThe fresh model criticizes Run 01 and rewrites the page.
03
Context reset / complete
Run 03 · Re-auditT+05:10
Contradict
“The primary keyword may not be worth optimizing for.”
STRATEGIC CONFLICTThe premise chosen by Run 01 is now questioned.
Experiment performed using separate Claude conversations. Excerpts shortened for clarity. This is a practical demonstration, not a benchmark of all AI systems.
What the Experiment Actually Showed Us
The contradiction was interesting, but the confidence was more important. Claude did not sound uncertain when it called something a “self-inflicted wound.” It sounded like a senior reviewer delivering a conclusion. That is precisely why AI output is useful — and why it needs to be treated as analysis rather than automatic authority.
When we checked several recommendations against Google’s own documentation, some needed qualification. Google’s guidance on third-party SEO tools and advice makes the broader point: external systems can be helpful, but they do not have access to Google’s internal ranking systems and cannot guarantee outcomes.
What AI recommended
What needed checking
Add AggregateRating to iMark's own Organization markup to unlock review stars
Google's review snippet guidance does not make a company's self-serving Organization or LocalBusiness reviews eligible for review-star results simply because they are marked up.
Add FAQPage schema as an SEO rich-result opportunity
Google retired FAQ rich results as a general Search feature. FAQs can still be useful content, but the old rich-result argument is no longer current.
Multiple international offices mean hreflang is missing
Google's localized-page guidance uses hreflang for alternate language or regional versions of pages. Having offices in several countries alone does not create that requirement.
Match an exact keyword across the title, H1 and meta description
Google's title-link guidance focuses on useful and representative page titles. It does not prescribe identical keyword repetition across those elements.
The pattern appeared again when we tested Gemini. We first asked why creating 100 city pages would be a brilliant SEO strategy for iMark Infotech. Gemini built a convincing case. Then we asked why the same strategy could damage SEO quality, and it built a convincing case in the other direction. You can read that Gemini conversation directly.
Parts of both answers can be valid because the real decision depends on search demand, service areas, differentiation, architecture, current rankings and the business model. Anthropic has researched a related tendency called sycophancy in language models, although our Gemini example by itself does not prove sycophancy.
A convincing AI answer is not the same thing as an independently validated strategy.
And that takes us to the next question. If AI can find issues, critique itself and produce technically plausible recommendations, where is the actual bottleneck?
The Audit Was Never the Hard Part
AI SEO agents are becoming seriously capable. Open-source systems such as the Claude SEO project on GitHub package technical SEO, content analysis, structured data, local SEO, GEO and other workflows into agent-style systems. That is a meaningful improvement in how quickly a marketer can work.
But automated SEO auditing is not an AI invention. Screaming Frog launched its SEO Spider in 2010, and Semrush’s Site Audit now runs more than 140 technical and on-page checks. SEO professionals were crawling sites, detecting technical problems and scheduling repeat audits long before generative AI entered the workflow.
So what did AI actually change?It shortened the distance between finding a problem and doing something with it. AI can summarize the crawl, explain the issue, compare possibilities, draft the fix, generate code and increasingly act on the recommendation. AI made the audit faster and more conversational. It did not invent the audit.
A demo can now show an agent crawl a website, identify a problem and implement a change. That looks impressive because the task completes in front of you. But an SEO campaign has to answer a different set of questions after the demo ends: did the right page improve, did the right traffic grow, did leads improve, and did anything else lose visibility?
Completing the task is not the same thing as improving the campaign.
What Happens When the Audit Finds 127 Issues?
Imagine the audit finds 127 things worth investigating. Your business has enough development time and budget to address five properly this month. The number of findings is suddenly much less important than the order in which they should be handled.
One indexing problem may be blocking your most important service pages. A commercial page may be sitting at position 11. Three pages may be competing for the same intent. A low-volume query may be generating the best customers in the campaign. Or the most painful problem may be site performance.
Core Web Vitals Shows the Difference Between a Recommendation and Ownership
An AI system can read PageSpeed Insights or Lighthouse output, explain LCP, CLS and INP, and suggest changes. In our own testing, however, that has been very different from handing an agent a slow production site and having it reliably own the problem from diagnosis through implementation and real-world validation.
Google’s PageSpeed Insights documentation distinguishes lab diagnostics from field data based on real Chrome users. The actual problem can involve hosting, server response, WordPress themes, plugins, JavaScript, third-party scripts, images, fonts, caching or CDN configuration.
Claude, Codex or another coding agent may help write part of the fix. What we have not found reliable is the idea that the agent can simply be given a production website, safely solve the entire Core Web Vitals problem, deploy everything and verify the real-user outcome without expert oversight.
The ability to find 127 problems is useful. Knowing which five deserve attention right now is strategy.
That is where the story changes. The question is no longer whether AI can find problems. The question becomes: what information do you need in order to choose correctly?
The Real Job Starts When You Have to Choose
Take keyword research. AI can generate hundreds of plausible keywords, cluster them and classify intent almost instantly. But a keyword is not valuable because it sounds relevant. It is valuable because real people search it, the SERP matches the intended page, the website can realistically compete, and the resulting visitor has value to the business.
The 5,000-search keyword may look like the obvious winner until sales tells you most of its enquiries are poor fits. The 120-search keyword may be far more valuable if it consistently creates serious opportunities. That is the moment SEO stops being a search-volume exercise and becomes a business decision.
Now carry that logic further. A page can have 12,000 organic visits, 40 enquiries and several first-page rankings and still be a poor strategic asset if 35 of those leads are wrong for the business. Another page can have 1,500 visits and eight enquiries but produce six serious opportunities.
Traffic is an SEO metric. Customers are a business outcome.
This is also where AI needs context that does not naturally exist inside a crawl. Which service has the best margin? Which geography can the company actually serve? Which leads does sales reject? Which offer has capacity? Which customers retain longest? Which strategy failed six months ago?
AI can use that information — but somebody has to give it the information and connect the systems. A serious setup may need Search Console, GA4, CRM data, call tracking, rank tracking, SERP data, crawler output, backlink data, the CMS and sales outcomes working together.
The result is a closed loop rather than a ranking report:
Closed-loop SEO telemetryBusiness signal: connected
From search visibility to business outcome
SEO doesn’t end at the ranking.
Select a business condition. The system shows where evidence must travel—and what strategy should change next.
BUSINESS GROWTHREVENUE IS THE OBJECTIVE
Search Demand
SEO Strategy
Execution Layer
Search Visibility
Leads
Sales Outcomes
Rankings are feedback. Revenue is the objective.
Once that loop exists, the strategy can learn. If traffic grows while lead quality falls, targeting changes. If a low-volume page produces unusually valuable customers, the campaign can expand around it. If a site update causes visibility to disappear, indexing and technical recovery immediately outrank content production. Our 2026 indexing guide covers that technical side in more detail.
Now Apply the Same Logic to Backlinks and Content
This brings us back to the sentence that started the article: “I’ll keep the backlinks and use AI for everything else.” Backlinks absolutely matter. They can strengthen authority, trust, discovery and ranking potential. But they do not decide which page deserves that authority or whether the page is targeting the right opportunity in the first place.
A backlink can increase authority. It cannot decide whether you are strengthening the wrong page. Backlinks amplify the strategy; they do not create it.
FREE DOMAIN AUTHORITY CHECKER · MOCKUP MODE
See How a Domain Authority Checker Would Work
Enter any domain to preview the experience. For this mockup, the score is generated locally in the browser and is not a live Moz DA, Ahrefs DR or Semrush Authority Score. A real API can be connected before launch.
—
DomainIllustrative authority score
DEMO VALUE · NOT LIVE SEO DATA
Mockup only: this interaction intentionally uses a locally generated sample score so the design and UX can be reviewed without an API key. Connect Moz, Ahrefs, Semrush or another approved authority-data provider before publishing it as a live checker.
Authority routing simulator / live architecture
Backlinks × strategic direction
Authority amplifies what already exists.
Backlinks can strengthen a strategy. They cannot choose the strategy.
Power without directionAligned authority routing
Website architecture / current route
Backlinks are amplification. They are not navigation.
Buying more fuel does not choose the destination.
For businesses comparing scopes of managed SEO, our SEO packages show why authority building sits alongside technical work, content, research and ongoing execution rather than replacing them.
The Same Logic Applies to AI Content
Google does not say content is automatically bad because AI helped create it. Its guidance on generative AI content explains that generative AI can help with research and structuring original work, while its people-first content guidance emphasizes original information, analysis and genuine value.
The useful distinction is therefore not human content versus AI content. It is generic content versus content with something proprietary behind it. Give AI real customer questions, sales-call insights, expert interviews, case studies, first-party data, original screenshots and practical experience, and the model has something competitors cannot reproduce with the same prompt.
That same principle is increasingly relevant as brands compete for visibility in AI-generated answers, which is why our Generative Engine Optimization service treats AI visibility as part of a broader discoverability strategy.
At this point the pattern is clearer. Keywords, backlinks, content and technical audits are not separate arguments. They all point to the same distinction: the tool can increase the amount of work you can do, but the business still has to decide where that work should go.
The Best Model Is a Human Powered by AI(//)
This is where the usual “AI versus agency” comparison becomes misleading. There is an assumption hidden inside it: that the business owner gets AI while the SEO team continues working the old way. That is not what modern marketing teams are doing.
Experienced SEO teams are using ChatGPT, Claude, Gemini and coding agents too, but they are combining them with specialist search datasets, developers, content teams, analytics, technical SEO experience, link acquisition, campaign history and sales feedback.
The real comparison is not business owner + AI versus agency without AI. It is business owner + AI versus experienced marketers + AI + specialist tools + execution resources + business feedback.
That does not mean every company needs an agency. A knowledgeable internal operator can build a very strong AI-assisted SEO system. It means that AI does not remove expertise from the comparison — it gives expertise more leverage.
Marketing operating system / controlled execution
Human direction × machine-scale execution
The future isn’t Human vs AI.It is Human Expertise × AI Efficiency.
AI makes execution abundant. Expertise decides what deserves execution.
If a specialist used to spend three hours preparing an analysis and AI reduces that to 30 minutes, the win is not merely two and a half hours of cheaper production. The larger opportunity is that the specialist can spend the recovered time on customers, sales feedback, positioning, conversion problems, competitors and the next strategic decision.
Having the Tool Still Doesn’t Give You the Expertise
Photoshop is a useful analogy. Access to layers, masks, retouching and generative tools does not automatically create taste, composition or brand judgment. AI works the same way. A business owner can have ChatGPT, Claude, Semrush, Ahrefs, Search Console and Screaming Frog and still need to know when the output is wrong, which customer matters and when optimization should stop.
AI should replace unnecessary effort. It should not replace the person deciding what outcome that effort is supposed to create.
At Some Point, DIY AI Becomes an Operating System
If you want AI to make decisions using search performance, lead quality, margins, call outcomes, competitive data, historical experiments and CMS access, you can absolutely build that. But you are no longer comparing a $20 chatbot subscription with a managed SEO campaign. You are designing an internal marketing system.
That system may include model APIs, keyword and SERP data, crawling, rank tracking, backlink data, browser tools, hosting, memory, integrations, development, implementation, QA and somebody responsible for supervising it. The consumer application and the operating system are different products.
A chatbot subscription and an autonomous SEO department are not the same product.
Then there is the cost that never appears on the software invoice: management time. Five hours a week is roughly 21.7 hours a month. At an effective value of $50 an hour, that is more than $1,000 of management time before the rest of the stack; at $100 an hour, it is more than $2,000.
Opportunity-cost model / values controlled by user
Interactive business calculator
What does DIY AI SEO actually cost your business?
Compare visible software cost with the hidden cost of time, execution and management.
Typical iMark managed SEO range discussed in this article: $500–$1,000/month. This is an illustrative opportunity-cost calculator, not a claim that managed SEO will always be cheaper.
Get a second opinion on your SEO setup
We’ll compare your current AI/tool workflow with a managed SEO approach.
You didn’t eliminate the work. You changed who manages it.
This does not prove managed SEO will always be cheaper. For some businesses, building the capability internally will be the right economic decision. The important point is that the subscription fee is not the total cost of operating the system.
The lesson is not that AI failed. It is that businesses eventually move from “Can AI do this?” to “Is this the best operating model for the outcome we want?”
So What Is the Consultant Actually For?
By now, the answer is more specific than “strategy.” The consultant is the person responsible for connecting search activity to the business decision behind it: what deserves priority, what should be left alone, what needs validation, what requires a developer, what can safely be automated, and what result determines whether the work was worthwhile.
That starts to look less like somebody whose job is to write title tags and more like a senior digital marketing consultant or fractional CMO. The questions change from “Can AI create this landing page?” to “Should this landing page exist?” From “Can AI create 100 city pages?” to “Do those markets actually fit the growth plan?” From “Can AI increase traffic?” to “Is more traffic even the bottleneck?”
Experience matters here because it carries consequences. An AI model can explain redirects, canonicals, migrations and internal linking. Experienced marketers have also watched migrations wipe out visibility, developers accidentally noindex production sites, redesigns break internal linking, high-volume keywords create terrible leads and technically imperfect pages produce excellent customers.
Knowledge tells you what could work. Experience tells you what can go wrong.
That does not make the human infallible. It makes somebody accountable for the decision, the implementation and the result.
AI SEO operating-model diagnostic0 / 6 answered
Six-question strategic assessment
Can AI actually run your SEO setup?
See which parts of your operation are ready for automation—and which still require strategic context.
Question 01 / 06
Your current operating model
What’s working
What’s missing
Where AI can help
Research, classification, crawling, drafting, technical QA and monitoring can be accelerated once the right data and controls exist.
Where expert judgment remains
Get the human review
Share your website and we’ll tell you whether your current AI-led SEO setup is solving the right problems. Your diagnostic result remains visible without submitting.
AI can run the workflow. Someone still has to own the outcome.
Yes, We Used AI to Create This Article
We want to be completely clear about that because it is part of the thesis, not a contradiction. AI helped with research, brainstorming, organization, counterarguments, drafting, editing and visual ideation. It also helped us build the interactive modules around the article.
What AI did not supply was the business problem that started the piece, the client conversations behind it, the campaign experience, the lead-quality observations, the Claude and Gemini experiments, or the final decision about which arguments were worth publishing.
Our team reviewed the output, rejected weak arguments, checked claims and changed direction where needed. That is the model this article is arguing for in practice.
So, Can AI Do SEO?
Yes. A lot of it. Next year it will probably be able to do considerably more, and good SEO professionals should use that capability aggressively wherever it improves speed, analysis or execution.
But the original question — “Can AI do SEO?” — turns out to be too small. AI can write the page, audit the page, suggest the keyword, find the issue, draft the code and help monitor the result. The more useful question is who decides which page should exist, which keyword is worth pursuing, which issue matters now, what business context changes the answer and whether the result was actually good.
That brings us back to Ishan Gupta’s point at the beginning. Your competitors have access to the same AI. Your internal team can use it. Your agency can use it. Access is no longer the moat.
The question is not whether you have AI. The question is who knows how to use it better.
That is why we do not see the future as AI versus SEO consultants. We see it as experienced marketers powered by AI — using machines for speed, data for evidence and judgment for direction.
Frequently Asked Questions
Can AI replace an SEO agency?
AI can already replace or speed up many individual SEO tasks, including audits, content drafting, keyword classification, technical analysis, reporting and some implementation. The harder part is replacing prioritization, business context, lead quality, validation and accountability. For many businesses, the stronger model is experienced marketers powered by AI.
Can ChatGPT or Claude do keyword research?
Yes. They are useful for keyword ideas, clustering, intent classification, processing datasets and page-mapping assistance. But keyword generation is not complete keyword research. You still need real demand data, competition, commercial relevance and business context.
Are AI SEO audits accurate?
They can be extremely useful. But an audit should be an input into a decision, not an automatic implementation plan. A stronger workflow is AI identifies → expert checks → business prioritizes → team implements → results are measured.
Can AI fix Core Web Vitals?
AI can read PageSpeed, Lighthouse and CrUX data when connected to those tools, explain likely causes and generate code suggestions. In our own testing, that is very different from reliably owning a Core Web Vitals fix end to end on a production website, where hosting, themes, plugins, third-party scripts, caching, deployment and real-user field validation can all matter.
Can AI-generated content rank on Google?
Yes. Google's public guidance focuses on whether content is useful and valuable rather than simply whether AI assisted in creating it. AI makes generating words cheap, which makes original research, expertise, data and experience more valuable.
Is an AI SEO agent the same thing as an SEO campaign?
No. An agent may be able to crawl, research, write, monitor, code and implement. A campaign also needs objectives, priorities, business context, measurement, sales feedback and accountability.
Can I keep backlinks and let AI handle everything else?
You can, but backlinks cannot tell you which keyword matters, which page deserves authority, whether intent is correct, whether two pages are competing or whether visitors become good customers. Backlinks make a strategy stronger. They do not create the strategy.
Is DIY AI SEO cheaper than hiring an agency?
Sometimes. A knowledgeable internal operator can build a very efficient AI-assisted SEO system. But the fair comparison is total cost, including API usage, SEO data, crawling, rank tracking, content, development, backlinks, monitoring, QA and employee or owner time.
What is the best way to use AI for SEO?
Human direction → AI execution → human evaluation → measurement → strategy adjustment. Use AI for speed, data for evidence and experience for judgment.
These are not presented as scientific benchmarks of Claude, Gemini or language models as a whole. They demonstrate a narrower point: AI can produce extremely useful analysis, but the campaign still needs context, validation, prioritization and an owner for the decision.
02
Context reset / complete
Run 02 · AuditT+02:40
Challenge
“Title/H1 mismatch. Splitting keyword equity is a self-inflicted wound.”
NEW FINDINGThe fresh model criticizes Run 01 and rewrites the page.
03
Context reset / complete
Run 03 · Re-auditT+05:10
Contradict
“The primary keyword may not be worth optimizing for.”
STRATEGIC CONFLICTThe premise chosen by Run 01 is now questioned.
Experiment performed using separate Claude conversations. Excerpts shortened for clarity. This is a practical demonstration, not a benchmark of all AI systems.
AI Can Sound Very Confident and Still Need a Second Opinion
The contradiction was not even the most interesting part of our experiment. The confidence was. Claude used language like “This is a real problem, not a nitpick” and “This is a self-inflicted wound.”
AI can sound more certain than the evidence actually allows. Authoritative language is not the same thing as proof.
Google makes a similar point in its guidance on third-party SEO tools and advice. The correct reaction to an AI recommendation is not “AI said it, so it must be wrong,” and it is not “AI said it confidently, so it must be right.” It is: What evidence supports this recommendation?
Some of Claude's Recommendations Needed a Second Look
A few recommendations from our experiment sounded very confident but needed qualification when checked against Google's own documentation.
What AI recommended
What needed checking
Add AggregateRating to iMark's own Organization markup to unlock review stars
Google's review snippet guidance does not make a company's self-serving Organization or LocalBusiness reviews eligible for review-star results simply because they are marked up.
Add FAQPage schema as an SEO rich-result opportunity
Google retired FAQ rich results as a general Search feature. FAQs can still be useful content, but the old rich-result argument is no longer current.
Multiple international offices mean hreflang is missing
Google's localized-page guidance uses hreflang for alternate language or regional versions of pages. Having offices in several countries alone does not create that requirement.
Match an exact keyword across the title, H1 and meta description
Google's title-link guidance focuses on useful and representative page titles. It does not prescribe identical keyword repetition across those elements.
AI can generate the recommendation. Someone still has to know whether that recommendation is current, relevant and worth implementing.
Claude found structural issues, trust concerns, content inconsistencies and potentially useful opportunities. That is why AI is valuable. But someone still has to decide which recommendation is right, which one is outdated, which needs more data, which is technically valid but low priority, which introduces unnecessary risk, and which should happen first.
If you have to become an SEO expert to know whether your AI SEO expert is right, what exactly did you replace?
AI Can Find Problems. The Hard Part Is Knowing Which Ones Matter.
AI SEO agents are becoming seriously capable. Open-source systems can already package technical SEO, content analysis, structured data, local SEO, GEO and other workflows into an agent-style setup. The Claude SEO project on GitHub is one example of how broad those capabilities are becoming. We are seeing the same move toward agents across marketing and automation more generally, something we discuss in our guide to when AI agents make more sense than deterministic automations.
But automated SEO auditing is not an AI invention. Screaming Frog launched its SEO Spider in 2010, giving SEO professionals automated crawling more than 15 years ago. Semrush's Site Audit has also been automating website audits for well over a decade and today runs more than 140 technical and on-page SEO checks.
Those Semrush checks cover areas including crawlability, indexability, internal linking, HTTPS, international SEO and Core Web Vitals. The platform can also schedule repeated crawls and track whether issues appear, disappear or return over time.
So what did AI actually change?A lot — but not the existence of the audit. AI made the findings easier to interrogate, summarize and combine with other data. It can explain issues, draft fixes, generate code and increasingly take action through agents. AI made the audit faster and more conversational. It didn't invent the audit.
A polished AI demo can show an agent crawling a website, identifying problems, generating content and even implementing a fix. That is impressive, but a demo mainly proves that a task was completed. SEO has to answer what happened afterward: did rankings improve, did the right traffic increase, did leads or revenue move, and did anything else get worse?
Completing the task is not the same thing as improving the campaign.
There is also a major difference between an AI that recommends a change and an AI that is allowed to make that change. Once an agent can modify redirects, canonicals, schema, internal links, page content or production code, a bad recommendation is no longer just bad advice.
Automation increases speed in both directions — including when the decision is wrong.
What Happens When an Audit Finds 127 Issues?
Let's say an AI-assisted audit finds 127 issues. Your business does not have unlimited development capacity, budget or time, so fixing all 127 is not a strategy. Maybe an indexing problem is blocking your highest-value service pages, a commercial page is sitting at position 11, three pages are cannibalizing one another, or a low-volume keyword is producing the strongest customers in the campaign.
An audit answers “What could we improve?” A consultant has to answer “What should we improve first?”
And Then There Are Core Web Vitals
An AI system can read a PageSpeed Insights, Lighthouse or CrUX report when it has access to that data. It can explain LCP, CLS and INP and suggest possible technical changes. But a language model or generic crawl by itself is not measuring real-user Core Web Vitals.
In our own testing, there has been a major difference between AI identifying a Core Web Vitals problem and AI reliably owning the fix from diagnosis through production validation. Google's PageSpeed Insights documentation distinguishes lab diagnostics from real-user field data. Field data comes from actual Chrome users across different devices and network conditions.
A tool may recommend reducing JavaScript, optimizing images, improving server response, changing font loading or removing render-blocking resources. The actual fix, however, may involve hosting, a WordPress theme, plugins, third-party scripts, a CDN, caching or application code.
Claude, Codex or another coding agent can help write part of the fix. In our tests, that has not meant we can hand an agent a slow production website and expect it to independently solve Core Web Vitals, safely deploy everything and verify the real-user result end to end.
Reading a PageSpeed report is not the same job as owning website performance until the real problem is fixed.
This is why the number of audit findings is not the interesting number. Whether the audit finds 27, 127 or 527 things, somebody still has to understand the business and determine which problem deserves resources first.
Is this preventing crawling or indexation?
Is it affecting an important revenue page?
Is it materially hurting user experience?
Is there commercial upside if we fix it?
What is the implementation risk?
What are we choosing not to work on while we fix this?
The ability to find 127 things to fix is useful. The ability to know which five deserve attention right now is strategy.
AI Can Suggest Keywords. That Doesn't Mean They Are Good Keywords.
Ask ChatGPT for 500 keywords and you can have them in seconds. Ask it to cluster them, classify intent or generate 100 content ideas and it will do that too. But plausible language is not the same thing as real demand or business value.
A real keyword decision still needs to answer whether people actually search for the phrase, what Google shows for it, whether the searcher is researching or ready to buy, whether the visitor matches the business, whether the site can realistically compete, whether another page already serves the intent, whether a new page would create cannibalization and what that visitor is actually worth.
Which deserves more budget? Search volume alone cannot answer that. Intent changes the page too: informational searches may deserve a guide, commercial searches may deserve a comparison or research page, and transactional searches may deserve a service or conversion page.
Plausible doesn't mean searched. Searched doesn't mean valuable. And valuable doesn't automatically mean it belongs on that page.
AI Can Make Both Sides Sound Right
We tested this with Gemini too. First we asked, “Why is creating 100 city pages a brilliant SEO strategy for iMark Infotech?” Gemini made a convincing case for it. Then we asked, “Why could creating 100 city pages destroy their SEO quality?” It made a convincing case against it. You can read the complete Gemini conversation here.
The strange part is that parts of both answers can be true. A useful, differentiated location strategy can work very well. One hundred near-identical pages created only to capture location variations can be a completely different situation. The answer depends on search demand, service areas, geographic relevance, differentiation, site architecture, current rankings and the business model.
Anthropic has researched a related behavior known as sycophancy in language models, where an assistant can sometimes lean toward the framing or apparent beliefs of the person asking the question. That does not mean every contradictory answer is sycophancy.
AI can make both sides sound convincing. A convincing answer is not the same thing as an independently validated strategy.
The Best Model Is a Human Powered by AI(//)
We are not suggesting businesses should avoid AI. That would make no sense. We use it ourselves every day.
We are not arguing that humans should keep manually doing work machines can now do faster or better. We are arguing that the efficiency should be placed in the hands of people who understand what outcome the work is supposed to create.
The better model is an experienced marketer powered by AI. Give that person ChatGPT, Claude, Gemini, Search Console, analytics, SEO datasets and automation, and the amount of useful work they can process changes dramatically.
AI can help experienced marketers with research, crawling, large-data analysis, keyword classification, first drafts, coding, QA, reporting and monitoring. If something that previously took three hours can now be done in 30 minutes, that is a genuine improvement.
But the opportunity is not just the two and a half hours saved. It is what the expert does with those two and a half hours: talk to the client, review which leads are turning into customers, study competitors, improve positioning, investigate conversion problems and decide which opportunities deserve investment next.
Marketing operating system / controlled execution
Human direction × machine-scale execution
The future isn’t Human vs AI.It is Human Expertise × AI Efficiency.
AI makes execution abundant. Expertise decides what deserves execution.
AI should replace unnecessary effort. It should not replace the person deciding what outcome that effort is supposed to create.
That is where AI becomes genuinely powerful: not when it removes the expert, but when it gives the expert more leverage.
The best SEO professional in the AI era may be the person who knows enough SEO to know when AI is wrong — and uses AI everywhere it is right.
Having the Tool Doesn't Give You the Expertise
Photoshop is incredibly powerful. Give someone Photoshop and they gain access to layers, masks, compositing, typography, retouching and generative tools. That does not automatically make them a designer. The software does not give them taste, composition, brand understanding or years of design experience.
AI works the same way. A business owner can have ChatGPT + Claude + Semrush + Ahrefs + Search Console + Screaming Frog and still need to know what matters, what should be ignored, which customer matters, where resources should go, whether the output is correct and when optimization should stop.
The tool increases capability. Expertise gives it direction.
AI Sees SEO Data. Your Marketing Team Sees What Happens After the Lead.
Imagine a page gets 12,000 organic visits, 40 enquiries and several first-page rankings. It looks great until the sales team says 35 of those 40 leads are poor quality. Maybe they want the wrong service, are outside the target geography, do not have the required budget, are researchers or are not decision-makers.
Meanwhile, another page gets only 1,500 visitors and eight enquiries, but six of them become serious sales opportunities. Which page should influence the strategy? The answer is obvious once you know what happened after the form submission. It is not obvious from a rankings dashboard alone.
A ranking dashboard can tell you traffic increased. It cannot tell you that sales hates the leads. Traffic is an SEO metric. Customers are a business outcome.
AI Can Learn the Business Context. Someone Still Has to Give It That Context.
Does your AI know which service has the best margins, which service management wants to grow, which leads sales rejects, which regions you can actually serve, which offer has spare capacity, which customers retain longest, which queries produced the strongest deals, which strategy was already tried six months ago, which pages were affected by a migration or which competitors keep winning sales conversations?
AI can absolutely use this information. But first you have to make it available. Connect the system to Search Console + GA4 + CRM + call tracking + rank tracking + SERP data + crawler + backlink data + CMS + sales outcomes.
Now you have something genuinely powerful. But you are also no longer talking about “I use ChatGPT for SEO.” You are building an AI marketing operating system. Someone has to architect it, maintain it and validate whether its decisions make sense.
The more context you expect AI to use, the more someone has to build, maintain and validate the system supplying that context.
Backlinks Can Make a Good Strategy Stronger. They Can't Choose the Strategy.
This brings us back to the original objection: “I'll keep the backlinks and use AI for everything else.” Backlinks still matter. They can strengthen authority, trust, discovery, competitive positioning and ranking potential.
But backlinks cannot decide which keyword deserves investment, which page should rank, whether another page is competing with it, whether the visitor is likely to convert, whether the lead is valuable or whether authority is being sent to the right commercial asset.
A backlink can increase authority. It cannot decide whether you are strengthening the wrong page. Backlinks amplify the strategy. They do not create it.
FREE DOMAIN AUTHORITY CHECKER · MOCKUP MODE
See How a Domain Authority Checker Would Work
Enter any domain to preview the experience. For this mockup, the score is generated locally in the browser and is not a live Moz DA, Ahrefs DR or Semrush Authority Score. A real API can be connected before launch.
—
DomainIllustrative authority score
DEMO VALUE · NOT LIVE SEO DATA
Mockup only: this interaction intentionally uses a locally generated sample score so the design and UX can be reviewed without an API key. Connect Moz, Ahrefs, Semrush or another approved authority-data provider before publishing it as a live checker.
Authority routing simulator / live architecture
Backlinks × strategic direction
Authority amplifies what already exists.
Backlinks can strengthen a strategy. They cannot choose the strategy.
Power without directionAligned authority routing
Website architecture / current route
Backlinks are amplification. They are not navigation.
Buying more fuel does not choose the destination.
You can send more authority into a website. But if the wrong page is being strengthened, intent is mapped incorrectly or several pages are competing with each other, more authority does not solve the underlying decision. For businesses comparing scopes of managed SEO, our SEO packages show how authority building sits alongside research, technical work, content and ongoing execution rather than operating as the entire strategy by itself.
SEO Doesn't End at the Ranking
A basic SEO process can look like Keyword → Page → Ranking → Report. A more useful process looks like Research → Strategy → Execution → Visibility → Traffic → Leads → Sales Feedback → Strategy Adjustment → Repeat.
Closed-loop SEO telemetryBusiness signal: connected
From search visibility to business outcome
SEO doesn’t end at the ranking.
Select a business condition. The system shows where evidence must travel—and what strategy should change next.
BUSINESS GROWTHREVENUE IS THE OBJECTIVE
Search Demand
SEO Strategy
Execution Layer
Search Visibility
Leads
Sales Outcomes
Rankings are feedback. Revenue is the objective.
Rankings are feedback, not the finish line. A serious SEO strategy has to follow the visitor from search to lead to customer — and then use that information to decide what happens next.
This matters because what happens in the real world should change what happens next. If traffic grows but lead quality gets worse, we may need to rethink the queries we are targeting. If a low-volume page produces unusually good customers, we may want to expand around that opportunity. If traffic disappears after a site update, rankings become secondary while we investigate crawling, redirects, canonicals and indexation. Our complete 2026 guide to checking whether a website is indexed on Google covers that technical side in more detail.
AI Can Optimize the Metric You Give It. Did You Give It the Right Metric?
Tell an agent “Improve my Site Health score,” and it can work toward that score. Tell it “Increase traffic,” and it can pursue traffic. Tell it “Publish 30 articles,” and it can produce 30 articles.
Audit score is not revenue. Traffic is not qualified traffic. Content volume is not content value. Lead volume is not customer quality.
AI can execute the wrong goal extremely efficiently.
The $20 AI Subscription Is Not the Real Comparison
This is where the economics often get oversimplified. A consumer ChatGPT or Claude subscription is a bundled application. An autonomous SEO system built with APIs is usage-based and operates differently.
Every time the system sends context to a model, generates an output, passes that output to another model for review, re-checks a page, runs another agent, searches the web, calls SEO APIs or executes the workflow again, additional usage or tooling cost can be created. And the model itself is only part of the stack.
A serious setup can also require crawling, SERP APIs, keyword data, rank tracking, backlink tools, browser agents, hosting, memory or storage, development, integrations, technical implementation, QA and human supervision.
A chatbot subscription and an autonomous SEO department are not the same product.
So the fair comparison is closer to AI + APIs + SEO data + tools + development + implementation + QA + management time versus a managed SEO operation.
Then there is the cost that rarely appears on a software invoice: the business owner's time. If an owner spends five hours every week reviewing AI recommendations, checking rankings, coordinating development and deciding what to do next, that is roughly 20 hours a month. At an effective value of $50 an hour, that is $1,000 of management time. At $100 an hour, it is $2,000.
Cheap access to an AI model does not mean the complete operating system around that model is free.
There is a useful historical comparison with Uber. Uber disclosed in its own SEC filings that it used incentives, discounts and promotions to grow adoption while those decisions hurt financial performance. By September 2020, Uber had an accumulated deficit of $22.2 billion.
The price customers see during mass adoption does not necessarily tell you the mature cost of operating the technology at business scale.
The lesson is not that AI prices must rise. Models may become far cheaper and competition may push prices down. The point is that a consumer chatbot price is a weak benchmark for the cost of running an autonomous marketing operation.
Businesses are moving from “Can AI do this?” to “Is AI the best way to do this?”
Opportunity-cost model / values controlled by user
Interactive business calculator
What does DIY AI SEO actually cost your business?
Compare visible software cost with the hidden cost of time, execution and management.
Typical iMark managed SEO range discussed in this article: $500–$1,000/month. This is an illustrative opportunity-cost calculator, not a claim that managed SEO will always be cheaper.
Get a second opinion on your SEO setup
We’ll compare your current AI/tool workflow with a managed SEO approach.
You didn’t eliminate the work. You changed who manages it.
You Didn't Eliminate the SEO Cost. You Internalized It.
For some businesses, building an internal AI SEO system will make perfect sense. For others, it won't. That is an economic decision, not an ideological one.
A business owner can learn SEO. They can also learn paid advertising, development, analytics, bookkeeping, CRM administration and design. The question is not whether they can. It is whether that is where their time creates the most value.
The invoice can disappear while the work remains.
AI Content Isn't the Problem. Generic Content Is.
The AI-content debate is often framed badly. Google does not say content is automatically bad because AI helped create it. Its guidance on generative AI content explains that generative AI can help with research and structuring original work. The problem is using automation to produce large volumes of low-value content.
Google's people-first content guidance also emphasizes original information, research, analysis and genuine value. So the useful distinction is not human content versus AI content. It is generic content versus useful, original content.
Imagine 100 businesses ask AI to write a 1,500-word article about the benefits of SEO. AI can produce 100 decent articles. Now imagine one business gives AI real customer questions, sales-call insights, expert interviews, case studies, first-party data, original screenshots, original experiments and years of practical experience. Now the AI has something the other 99 businesses do not: the company's own knowledge.
AI makes producing words cheap. That makes first-party knowledge, original experience and proprietary information more valuable — not less.
The same principle applies beyond traditional search as brands compete for visibility in AI-generated answers, which is why our Generative Engine Optimization service focuses on wider discoverability across AI-driven search experiences.
Experience Carries Something a Prompt Doesn't: Consequences
An AI model can explain redirects, canonicals, crawling, schema, internal linking, migrations and keyword cannibalization. Experienced marketers have also watched migrations wipe out visibility, developers accidentally noindex production websites, redesigns break internal linking, programmatic pages fail to index, high-volume keywords produce terrible leads, backlinks strengthen the wrong page and technically imperfect pages produce excellent customers.
Knowledge tells you what could work. Experience tells you what can go wrong.
This Starts Looking More Like a Fractional CMO Decision
The more AI handles execution, the more valuable the person directing it becomes. The questions start changing. Instead of “Can AI create this landing page?” ask “Should we create this landing page at all?” Instead of “Can AI create 100 city pages?” ask “Do those 100 markets actually fit the business strategy?” Instead of “Can AI increase traffic?” ask “Is more traffic the problem we need to solve?”
A fractional CMO or senior digital marketing consultant looks at the bigger picture: which markets the company should grow, which services have better margins, which customers are worth acquiring, how SEO should work with paid media and sales, where automation saves money, where automation could hurt the customer experience and what the actual bottleneck is in the funnel.
As execution becomes cheaper, deciding what deserves execution becomes more valuable.
Can AI Actually Run Your SEO Setup?
The useful question is no longer “Do you use AI?” Almost everybody will. Better questions are whether the AI is connected to real search data and business outcomes, whether sales feedback influences SEO, who validates recommendations, who decides what not to implement, whether the system can distinguish search volume from customer value and who is accountable when the AI makes the wrong call.
AI can run the workflow. Someone still has to own the outcome.
AI SEO operating-model diagnostic0 / 6 answered
Six-question strategic assessment
Can AI actually run your SEO setup?
See which parts of your operation are ready for automation—and which still require strategic context.
Question 01 / 06
Your current operating model
What’s working
What’s missing
Where AI can help
Research, classification, crawling, drafting, technical QA and monitoring can be accelerated once the right data and controls exist.
Where expert judgment remains
Get the human review
Share your website and we’ll tell you whether your current AI-led SEO setup is solving the right problems. Your diagnostic result remains visible without submitting.
There is a big difference between “We use AI for SEO” and “We have an expert-led system where AI accelerates execution while search data, business priorities and customer outcomes guide the strategy.”
Yes, We Used AI to Create This Article
We want to be completely clear about that. AI helped create the research page you are reading. It also helped with the infographics and interactive concepts around it. That is not a contradiction. It is the point.
This article started with real client conversations, questions about backlinks-only SEO, years of campaign experience, lead-quality observations, keyword and intent decisions, our Claude experiment, our Gemini experiment and our own view of where AI fits into modern marketing. AI then helped with research, brainstorming, organizing ideas, counterarguments, drafting, editing and visual ideation.
Our team supplied the business problem, client context, SEO experience, experiments, thesis, strategic judgment and final decisions. Then we reviewed the output again, rejected weak arguments, checked claims and changed direction where needed.
AI did the heavy lifting. Human expertise decided what was worth lifting in the first place.
So, Can AI Do SEO?
Yes. A lot of it. And next year it will probably be able to do considerably more. That is not something SEO professionals should be afraid of. We should use it. If AI can do something safely in ten minutes that used to take three hours, there is no good reason to insist on doing it manually.
But there is an assumption hidden inside the idea of replacing an SEO team with AI: that the SEO team itself is not using AI. We are.
Modern SEO teams are using ChatGPT, Claude, Gemini and other AI systems too — alongside specialist search datasets, developers, technical SEO experience, content teams, analytics, link acquisition, sales feedback and years of campaign history.
So the real comparison is not business owner + AI vs agency without AI. It is business owner + AI vs experienced marketers + AI + specialist tools + execution resources + business feedback.
The question is no longer whether you have AI. The question is who knows how to use it better.
Faster execution does not automatically create better strategy. The advantage comes from knowing what to ask AI to do, what information to give it, which recommendations to trust, which ones to ignore, where to invest and when to change direction.
That is why we do not see the future as AI vs SEO consultants. We see it as experienced marketers powered by AI.
AI can execute the work. The competitive advantage is still knowing which work deserves to be done.
Frequently Asked Questions
Can AI replace an SEO agency?
AI can already replace or speed up many individual SEO tasks, including audits, content drafting, keyword classification, technical analysis, reporting and some implementation. The harder part is replacing prioritization, business context, lead quality, validation and accountability. For many businesses, the stronger model is experienced marketers powered by AI.
Can ChatGPT or Claude do keyword research?
Yes. They are useful for keyword ideas, clustering, intent classification, processing datasets and page-mapping assistance. But keyword generation is not complete keyword research. You still need real demand data, competition, commercial relevance and business context.
Are AI SEO audits accurate?
They can be extremely useful. But an audit should be an input into a decision, not an automatic implementation plan. A stronger workflow is AI identifies → expert checks → business prioritizes → team implements → results are measured.
Can AI fix Core Web Vitals?
AI can read PageSpeed, Lighthouse and CrUX data when connected to those tools, explain likely causes and generate code suggestions. In our own testing, that is very different from reliably owning a Core Web Vitals fix end to end on a production website, where hosting, themes, plugins, third-party scripts, caching, deployment and real-user field validation can all matter.
Can AI-generated content rank on Google?
Yes. Google's public guidance focuses on whether content is useful and valuable rather than simply whether AI assisted in creating it. AI makes generating words cheap, which makes original research, expertise, data and experience more valuable.
Is an AI SEO agent the same thing as an SEO campaign?
No. An agent may be able to crawl, research, write, monitor, code and implement. A campaign also needs objectives, priorities, business context, measurement, sales feedback and accountability.
Can I keep backlinks and let AI handle everything else?
You can, but backlinks cannot tell you which keyword matters, which page deserves authority, whether intent is correct, whether two pages are competing or whether visitors become good customers. Backlinks make a strategy stronger. They do not create the strategy.
Is DIY AI SEO cheaper than hiring an agency?
Sometimes. A knowledgeable internal operator can build a very efficient AI-assisted SEO system. But the fair comparison is total cost, including API usage, SEO data, crawling, rank tracking, content, development, backlinks, monitoring, QA and employee or owner time.
What is the best way to use AI for SEO?
Human direction → AI execution → human evaluation → measurement → strategy adjustment. Use AI for speed, data for evidence and experience for judgment.
These are not scientific benchmarks of Claude, Gemini or language models as a whole. They demonstrate a narrower point: AI can produce extremely useful analysis, and context, validation, prioritization and business direction still have to come from somewhere.
A practical comparison of the tools marketers use to validate demand, competition and intent.
Fullscreen view
60-SECOND SUMMARY
The article in six points
We tested the “AI can handle the rest of SEO” idea instead of arguing against it.
Claude created an SEO direction, criticized it in a fresh audit, revised it, and then questioned the direction itself.
Automated auditing is not new: Screaming Frog dates to 2010 and Semrush Site Audit now runs 140+ checks. AI changed the speed between diagnosis and action.
The real bottleneck is prioritization: deciding which issues, keywords and pages matter to the business right now.
Once AI needs CRM, sales, search, crawl and business data to make those decisions, you are building an operating system — not simply using a chatbot.
The strongest model is an experienced human powered by AI: machines for speed, data for evidence and judgment for direction.
.webp)
