Is your AI giving bad SEO advice? Do you know the difference?

What happens when an AI follows your SEO instructions—and still builds the wrong thing?

I recently audited a newly launched website with several avoidable SEO problems.

The client was genuinely surprised. They had not ignored SEO. Their response after seeing the audit was telling:

"I can't believe all of these issues are here. I included instructions in my prompt to build the site with SEO best practices."

The prompt had been followed. The problem was that the instruction itself did not offer the protection they thought it did.

That raises a more important question than whether an AI can provide SEO recommendations:

How would you know whether its advice was good, incomplete, or quietly harmful?

“Build this using SEO best practices” sounds like a reasonable instruction.

An LLM can produce clean code, structured copy, familiar recommendations and confident explanations. It can generate an initial solution much faster than someone starting from a blank page.

The problem is not that AI knows nothing about SEO.

The problem is that the prompt assumes there is one agreed-upon set of best practices. It assumes the model understands which practices apply to this website, this technical stack, this business and this market.

It assumes the prompt contains enough context. It also assumes someone will review the result who can recognize what is missing.

That is a much larger assumption than it appears.

Even official search documentation does not present SEO as one universal checklist. Google’s guidance is divided across crawling, indexing, JavaScript, canonicalization, internal linking, structured data, content quality and other areas because different systems create different risks. Google also makes clear that following general guidance does not guarantee that a page will be indexed.

“Use SEO best practices” does not tell a model how the site makes money, which pages matter most, which content must remain accessible without client-side rendering, how duplicate URLs should be handled or what trade-offs the business is willing to accept.

AI is a multiplier. It can accelerate good judgment. It can also scale gaps, assumptions and technical debt.

The most obvious limitation is also the hardest to solve. You may not know which questions to ask. You may not know which constraints matter. You may not know which templates require special handling, which recommendations are technically correct but wrong for the business or which important requirements the model failed to mention entirely.

A prompt can only include the context the user knows to provide. This creates a circular problem:

You need SEO knowledge to ask for reliable SEO guidance. You also need SEO knowledge to evaluate the answer.

The fluent output hides that gap. A recommendation can sound complete without being complete. It can explain five things clearly while omitting the sixth thing that determines whether the implementation works.

Consider JavaScript rendering. A model may correctly produce a functional client-side experience. The pages may work perfectly for users in a browser. But that does not settle whether important content and links are available to search systems in the right form.

Google can process JavaScript, but its own documentation notes that JavaScript sites introduce differences and limitations that developers need to account for during crawling, rendering and indexing. It also describes dynamic rendering as a workaround rather than a recommended long-term solution because of the extra complexity it creates.

The site can work.

The code can look clean.

The recommendation can sound reasonable.

The SEO implementation can still be wrong.

It would be easy to treat this as a problem for people who know very little about SEO.

It is not.

Experienced SEOs have blind spots too.

A technical SEO may quickly recognize crawling, indexing, rendering and canonicalization risks. That same person may be less qualified to judge whether an AI-generated content strategy reflects the customer, the market or the brand.

A content strategist may be excellent at evaluating search intent and messaging but less likely to notice that internal links are not rendered as crawlable links.

An international SEO specialist may recognize localization and market-selection problems that a generalist would never think to include in the prompt.

Each part of SEO carries its own hidden context. That means the risk is not simply AI versus expert.

It is often AI plus a specialist operating outside their strongest area.

That combination can be particularly dangerous because some of the output will be correct. Familiar terminology and technically accurate observations create confidence that spills into areas where neither the user nor the model has enough context. The problem is not ignorance alone.

It is partial knowledge presented as complete judgment.

None of this means AI-generated SEO guidance is useless. AI can create a strong initial checklist. It can explain an unfamiliar concept in plain language. It can challenge an existing plan, identify areas for further investigation and turn a known process into a repeatable workflow.

It can help specialists move faster within areas they already understand.

It can also help teams widen their initial field of view. Someone focused on content might ask a model to identify possible technical dependencies. A technical SEO might use it to generate questions for a content or brand review.

That is useful.

Official search guidance takes a similarly conditional position on generative AI. Google says it can help with research and content structure, while warning that generating pages at scale without adding value may violate its spam policies. The distinction is not whether AI was involved. The distinction is what the resulting system produces for users. The same principle applies to SEO advice.

AI is useful as a second set of eyes. It is risky as the final authority.

Its strongest role is often to expand the questions under consideration, not close the decision.

Human judgment remains the bottleneck.

Not because humans are always right. They are not.

But someone still needs to understand the system, test the recommendation, weigh the trade-offs and remain accountable for the result.

Someone must decide what happens to crawling, indexing, rendering, signals and trust.

The wrong question is: “Can AI give me SEO advice?”

Of course it can. The better question is: “What role should this advice play in the decision?”

AI can generate possibilities. It should not grant final approval. Ask it to state its assumptions. Ask what information is missing. Ask which recommendations depend on the rendering model, content type, market, business objective or implementation scale.

Separate technical questions from content questions. Separate authority questions from business questions. A correct answer in one area does not validate the rest.

High-impact decisions should still be reviewed by people with relevant expertise. Recommendations should be checked against the actual website, not accepted because they resemble a generic best-practice list.

That changes accountability. “An AI recommended it” is not a rationale.

“We tested the implementation, confirmed the assumptions and accepted the trade-off for these reasons” is. The goal is not to remove AI from the process.

The goal is to stop treating confident language as evidence of reliable judgment.

Your AI may be giving you good SEO advice.

It may also be giving you generic advice, incomplete advice or advice that becomes harmful when applied across an entire site.

Those outputs can look remarkably similar. The difficult part is not getting an answer.

It is knowing the difference before the consequences appear.

Model Parameters: Internal values learned during training that shape model behavior.

Learn More

One more thing: AI is only as good as its operator, and if you are reading this newsletter, you’re better than most!

Till next time,

PS: Let me know what you think of this issue, or anything else here: [email protected]