Introduction
This post is for two readers. The first is the marketer or in-house SEO who keeps receiving forwarded ChatGPT screenshots from a client with the subject line “Why aren't we doing this?” The second is the small business owner — possibly the same person sending those screenshots — who is trying to figure out whether the SEO advice they just received from an outside vendor is good guidance or AI-generated filler dressed up to look like expertise.
Both readers are in a hard spot. The marketer cannot dismiss ChatGPT output without sounding defensive. The owner cannot evaluate a vendor pitch without some kind of working framework. And both readers are operating in a moment when ChatGPT and Claude have become the default way many people try to learn anything new — including SEO — and when the volume of confident, plausible-sounding, partially wrong AI-generated advice in the wild has gone from “occasional” to “constant.”
The Search Engine Land community has been wrestling with this directly. In a piece published May 14, 2026 by Frank Olivo, Olivo argues against the instinct to argue back: “The key isn't arguing with AI outputs. It's guiding stakeholders through what's useful, generic, or simply incorrect.” That is a more honest framing than most agencies use, and it is the right starting place for this playbook.
Key Takeaways
- ChatGPT SEO advice fails in three predictable ways: outdated training data, generic one-size-fits-all recommendations, and confident inaccuracy when the model does not know what it does not know.
- The response template Olivo recommends starts with validating the effort, leads with what is genuinely worth exploring, lets the sender reach their own conclusions, and reframes the issue as a prompt-and-context problem rather than an AI failure.
- For SMB owners receiving a vendor pitch, the diagnostic is the same shape in reverse: is the advice specific, sourced, falsifiable, and proportional to the claimed payoff?
- Three red flags in any pitch — surface-level audit findings, generic best-practice language without site-specific tying, claims of results without disclosed methodology.
- We are honest about a limitation: this entire piece is about honesty, and our own recommendations are framed as “we recommend” rather than presented as universal facts.
Why Does ChatGPT SEO Advice So Often Sound Right and End Up Wrong?
The trap is not that ChatGPT is bad at SEO. The trap is that ChatGPT is fluent at SEO — fluent enough to compose advice that reads like a confident professional wrote it, with no visible markers that distinguish the parts the model knows from the parts the model is improvising.
Three failure modes account for most of the cases we see in the wild.
Outdated training data. Most LLMs have a training cutoff and a thinner layer of more recent retrieval on top of it. A 2026 question about Google's spam policies, AI Overviews citation behavior, or the recent core update can get answered with a pattern the model learned from a 2022 article. The answer can be precise, fluent, and obsolete. The reader has no signal that the underlying source is years old unless the model volunteers it — which it often does not.
Generic genericity. ChatGPT defaults to the safest, most general advice the training data supports. “Build high-quality backlinks.” “Optimize your title tags.” “Create comprehensive content.” These statements are not false; they are also not actionable for a specific site in a specific industry in a specific market. Olivo describes this directly: the AI produces “generic word count recommendations (3000+ words) contradicted by top-ranking results” and “mischaracterizes competitor focus strategies without website verification.”
Confident inaccuracy. This is the hardest failure mode because the surface signal — fluent prose, plausible structure, no hedging — is the same signal humans use to assess expertise in other humans. When a model does not know something, it does not always know that it does not know, and it does not always tell you. Reporting on this dynamic is well-established. MIT Technology Review's ongoing AI coverage and the work coming out of Stanford HAI have documented the gap between LLM fluency and LLM accuracy as a persistent feature, not a bug that has been fully solved. Anthropic's own core views on AI safety treats this as an open research problem rather than a finished one.
For an SMB owner reading a ChatGPT response about their own SEO, the practical effect is that the wrong advice is indistinguishable from the right advice on the surface. That is the whole reason a response framework matters.
What Is the Right Response When a Client Forwards ChatGPT SEO Advice?
Olivo's four-step framework is the cleanest version we have seen, and it maps directly onto what works in our own client conversations:
- Validate the effort. Start with “thanks for sending this over.” The client took initiative; the relationship rests on that being respected. Olivo's reasoning: “Most people forwarding ChatGPT output think they are being helpful... If your first move is to attack the recommendations, they will hear you attacking the effort.”
- Lead with what is worth exploring. Find the genuinely useful observation in the AI output and acknowledge it first. There almost always is one — a topic the client should have on their radar, a question worth asking, a tactic that is reasonable in context. Naming it first changes the dynamic of the entire conversation.
- Let the sender reach their own conclusions. Rather than declaring the AI advice wrong, present the contradictory evidence calmly. “ChatGPT recommended 3,000-word posts. Here are the top three ranking pages for that query. Notice the average length is under 1,200 words. What do you think is happening here?”
- Reframe as a prompt and context problem. Olivo's line: “AI outputs are only as good as the prompt and context they are given.” This is the gracious exit ramp that lets the client see the limitation without feeling that they personally got fooled.
The whole framework is built around a single principle: “Don't debate ChatGPT. Show the person who sent the recommendations that you can evaluate AI output objectively and professionally.”
There is a follow-on move we recommend layering on top of Olivo's framework, especially for ongoing client relationships: proactively share what good AI-assisted SEO research looks like. If you can show the client how you would have asked the question — with the site context, the industry constraints, the specific competitors loaded into the prompt — you have demonstrated competence and given them a tool, not just a correction. This is the same posture we walked through in what blog posts get cited in ChatGPT. The agencies that win in 2026 are the ones that treat AI as a collaborator they out-prompt, not a competitor they have to defend against.
The 8-Question Diagnostic for Evaluating Any AI-Generated SEO Recommendation
This is the connective tissue between the marketer audience and the SMB-owner audience. Both can run the same diagnostic on a piece of AI-generated SEO advice. The questions are designed to be answerable in 10 minutes; nothing here requires deep technical expertise.
- Is this advice specific to my site, industry, or market? If the same recommendation could be pasted under any business name and still read fluently, it is generic. Generic is a yellow flag, not necessarily a red one — but it changes how much weight the recommendation deserves.
- Does the source cite anything published in the last 12 months? SEO changed materially between 2023 and 2026. Advice that does not reference recent guidance, recent updates, or recent SERP behavior should be treated with caution.
- Is the recommendation falsifiable? Can you measure whether it worked? “Improve E-E-A-T” is not falsifiable. “Add author bylines and last-updated dates to your top 10 landing pages and track the impact in Search Console over 30 days” is.
- Is it consistent with Google's official guidance? Google's Search Central documentation is the canonical source for what Google says it wants. Recommendations that contradict that documentation should clear a higher bar.
- Does it pass the “ask three SEO professionals” gut-check? Would three working SEO professionals — not the model — agree this is current best practice? If you do not personally have three SEO professionals on speed-dial, the answer to this question is itself a clue: maybe the next step is finding one.
- Is the proposed work proportional to the expected payoff? A recommendation that asks for a six-month content sprint and promises “improved rankings” is not proportional. A recommendation that asks for a one-day technical fix and promises a measurable lift in a specific Search Console metric is.
- Does it ignore obvious downsides? Real SEO recommendations come with trade-offs. AI-generated recommendations often present an unalloyed positive. If there is no honest acknowledgment of cost, risk, or competing priorities, that absence is itself a signal.
- Does it sound like a generic checklist or a custom diagnosis? This is the gut-level version of question 1. A checklist treats your site as interchangeable with every other site in your industry. A diagnosis treats your site as a specific entity with specific problems. The two read very differently if you are looking for the distinction.
A recommendation does not have to pass all eight questions. But the more it fails, the more carefully it should be evaluated before any work is committed.
The Honest-Response Template for Marketers
Here is the script we use internally at Button Block when a client forwards ChatGPT SEO advice. It is meant to be edited for voice; the bones are the part that matters.
“Thanks for sending this — it is a useful prompt to look at [topic] together. There are two things in this response that are worth exploring: [genuine observation #1] and [genuine observation #2]. On the rest, a couple of nuances worth flagging:
[Specific issue 1 — describe the concrete reason this recommendation does not fit the site, with one piece of evidence. Example: ‘The 3,000-word recommendation runs counter to what we see in the SERPs for your queries — the top three results average closer to 1,100 words and rank on focused, specific content rather than length.’]
[Specific issue 2 — same structure, different angle. Example: ‘The backlink target the model cited assumes a kind of outreach scale that does not fit your team size or your brand voice. We have a leaner approach that fits the same goal.’]
One general note: tools like ChatGPT are useful for surfacing questions, but the answers depend heavily on the context you load into the prompt. We can pair on a session where we run your specific site context through one of these tools together and see what comes back — that is often a more productive use of the AI than the default ungrounded query.”
What this script does:
- Validates the effort before introducing any correction (Olivo's step 1).
- Names a genuine observation worth exploring before any pushback (step 2).
- Lets the contradicting evidence do the work rather than the marketer's authority (step 3).
- Reframes the issue as a prompt-and-context problem with a constructive next step (step 4).
- Closes with a collaborative invitation rather than a competitive boundary.
In our experience, this version of the conversation almost always lands. The version that does not land is the one where the marketer leads with “actually, ChatGPT is wrong” — even when the marketer is right.
The Honest-Evaluation Template for SMB Owners
If you are the small business owner reading a vendor pitch email — or a strategy document, or a “discovery audit” report — and trying to figure out whether it is good advice or AI-generated filler, here is the version that works in reverse.
Three red flags:
- Surface-level audit findings. A real audit refers to specific pages, specific queries, specific Search Console data, specific competitors. An AI-generated audit refers to “your site,” “your industry,” and “your competitors” in the abstract. If the pitch never names anything specific about your business, that is a red flag.
- Generic best-practice language without site-specific tying. “Focus on long-form, comprehensive content” is best-practice language. “Your service-area pages for [specific town] are thin and your competitors have substantially more depth — here is what we would add” is site-specific. If the pitch is mostly the former, ask for the latter before committing.
- Claims of results without disclosed methodology. “We can get you to page one of Google” is a claim. “Here is the keyword set, here is the current rank, here is the realistic 90-day movement based on the competitive landscape, and here is what could go wrong” is a methodology. Pitches that promise results without showing their work are pitches written for someone who will not read closely.
Three green flags:
- Specific examples from your actual site. The vendor has looked at your site before sending the pitch and can point to specific pages, specific structural issues, or specific opportunities they noticed. This takes longer to produce than a templated email, and that is the whole signal.
- Transparent about uncertainty. The pitch acknowledges what it does not know, what would require deeper investigation, and what the realistic range of outcomes is. AI-generated pitches almost never do this; humans who have actually done the work almost always do.
- Distinguishes facts from recommendations. Sourced claims are presented as sourced (“according to Google's documentation,” “according to the recent Search Engine Land analysis”). Opinions are presented as opinions (“we recommend,” “in our experience”). The two are not mixed.
We covered the broader trust dynamics in AI search reputation management for small business — the short version is that AI-generated noise raises the value of vendor relationships that demonstrably show their work, and lowers the value of vendor relationships that do not.
There is an adjacent failure mode worth flagging directly: SEO advice that looks correct in the SERPs but does not show up in AI search. As we covered in why your content doesn't appear in AI Overviews and how AI Overviews are reshaping paid search, the two surfaces are different ecosystems with different signals. A vendor pitch that ignores the AI search layer entirely is increasingly missing half the picture.
Three Fort Wayne SMB Scenarios
These are composite scenarios drawn from the kinds of conversations we have with Fort Wayne, Auburn, and New Haven small businesses. None describes a specific client.
A Fort Wayne HVAC owner gets a cold email. The email says, “Our analysis shows you need 100 backlinks this month to rank for ‘HVAC repair Fort Wayne.’” The owner asks us if this is real. The 8-question diagnostic reads it instantly: not specific to the actual site, no recent source cited, not falsifiable (what counts as a “needed” backlink?), inconsistent with Google's longstanding guidance against link buying, and not proportional (100 backlinks in 30 days is a pattern that triggers Google's spam systems, not one that builds rankings). The right move is to politely decline and to take the prompt as a signal — if a cold AI-generated email got this far, the owner's site probably is competing in a contested local pack and a real conversation about local SEO is overdue.
An Auburn boutique owner reviews their Squarespace developer's updates. The developer has pasted ChatGPT-generated meta descriptions across the entire product catalog. The boutique owner is not technical and is not sure if this is good or bad. The diagnostic is gentle: are the descriptions specific to each product, or are they templated? Do they include the product's actual differentiators? Do they read like sentences a human in Auburn would write, or like AI-generated boutique-store boilerplate? If the answer is “templated and boilerplate,” the work is not bad enough to undo but it is not good enough to keep — the right next step is a session where the owner edits the top-10 product descriptions in her own voice and the rest get treated as placeholders.
A New Haven law firm partner asks ChatGPT for SEO advice. The model recommends “create lots of blog content about personal injury keywords.” The partner is rightly skeptical. The diagnostic flags the recommendation: not specific (every personal injury firm gets this advice), not aligned with what works in the legal vertical (depth and authority matter more than volume), and ignores the YMYL implications of the legal category entirely. The right conversation for this firm is not about more blog content; it is about establishing demonstrable authority in a specific practice area in a specific geography — the same posture we walked through in topical authority is not enough for AI search and the broader local AEO logic in the 7 AEO tools Fort Wayne small businesses need.
What unites all three scenarios is the same observation: the AI-generated advice was not maliciously wrong. It was generically right and specifically useless. That is the failure mode the diagnostic is built for.
A Note on What This Post Is Not
This is not a “ChatGPT is bad” post. We use ChatGPT, Claude, and Perplexity every day at Button Block — for research, for first-draft writing, for prompt-testing AI search visibility, for surfacing questions we should be asking. The point of this post is not to push back on AI; it is to push back on AI-generated advice that gets treated as expert advice without the diagnostic that separates the two.
The Search Engine Land coverage on the same theme is worth reading directly. The piece on why good content still loses covers the SERP side, the study on ChatGPT citations covers what AI search actually rewards (and we wrote our own take in ChatGPT citations favor ranking and precision), and the piece on AI search as a reputation risk covers the downstream effect when the wrong things get said about your business in AI answers. Together they are a more honest map of the territory than any single ChatGPT response will give you.
For the broader policy context on how the models themselves are governed, OpenAI's usage policies and Anthropic's published positions are the primary sources worth bookmarking.
Want an Honest SEO Diagnosis, Not a Generic AI Checklist?
If you have a stack of AI-generated SEO recommendations on your desk and you cannot tell which ones are worth acting on, that is the moment a real conversation pays for itself. Button Block builds custom diagnostic engagements for Fort Wayne and Northeast Indiana small businesses — the deliverable is a specific action list built from your actual site, your actual market, and your actual competitive landscape.
Ready for an Honest SEO Conversation?
Explore our SEO Services or reach out for a no-pressure intro call. We will tell you honestly what is worth doing, what is not, and what falls in the middle.
Frequently Asked Questions
- Why is ChatGPT often wrong about SEO?
- ChatGPT and similar large language models are trained on a snapshot of web content with a cutoff date, then layered with a thinner retrieval system for more recent information. SEO is a fast-moving field where Google's guidance, AI Overviews behavior, and SERP dynamics shift materially every quarter. The result is that ChatGPT can give fluent, confident, and out-of-date advice with no visible signal that the underlying source is years old. Stanford HAI and MIT Technology Review have both documented this as a persistent feature of LLM behavior, not a fully solved problem.
- How should I respond to a client who sends me ChatGPT SEO recommendations?
- The Search Engine Land framework from Frank Olivo is the cleanest version: validate the effort first, lead with what is genuinely worth exploring, present contradicting evidence calmly rather than declaring the AI wrong, and reframe the issue as a prompt-and-context problem. The principle Olivo emphasizes is "Don't debate ChatGPT. Show the person who sent the recommendations that you can evaluate AI output objectively and professionally."
- How can a small business owner tell if a vendor pitch was written by AI?
- Three red flags: surface-level audit findings with no specific reference to your actual pages, generic best-practice language without any site-specific tying, and claims of results without a disclosed methodology. Three green flags: specific examples from your real site, transparent acknowledgment of uncertainty, and a clear distinction between sourced facts and the vendor’s recommendations. The pitches that show their work are almost always written by humans who have actually done the work; the pitches that do not, often are not.
- Is it ever appropriate to follow SEO advice that ChatGPT generated?
- Yes, when the advice passes a real diagnostic. The 8-question test in this post is designed to be runnable in 10 minutes. Recommendations that are specific to your site, sourced to recent guidance, falsifiable, consistent with Google's official documentation, proportional to the claimed payoff, and honest about trade-offs are worth acting on regardless of who or what wrote them. The diagnostic is the filter, not the source.
- What is the single most common mistake when responding to AI-generated SEO advice?
- Leading with the correction instead of with validation. When a marketer’s first move is "actually, ChatGPT is wrong," the client hears it as an attack on their initiative, not on the AI. The conversation breaks down before the substantive disagreement gets aired. Olivo’s framework and our own experience both point to the same fix: acknowledge the effort, find the genuine observation worth exploring, then introduce the nuance.
- How do I prevent ChatGPT SEO advice from becoming a recurring problem with my clients?
- The most effective move is proactive: share what good AI-assisted SEO research looks like, with the site context, industry constraints, and specific competitors loaded into the prompt. When a client sees how you would have asked the question, the dynamic shifts from "we are debating the AI" to "we are using the AI together, more skillfully." That dynamic is more durable than any single response template.
- How should Fort Wayne small businesses evaluate ChatGPT SEO advice in 2026?
- The same diagnostic applies regardless of geography, but the local layer matters. For a Fort Wayne, Auburn, or Allen County small business, the highest-value question to add to the 8-question test is: "Does this recommendation actually account for the competitive landscape and search demand in Northeast Indiana?" Generic ChatGPT advice almost never does. We recommend pairing any AI-generated audit with a human review tied to the actual NE Indiana market before committing budget.
Sources & Further Reading
- Search Engine Land: Your client sent ChatGPT SEO advice: Here's how to respond — Frank Olivo's four-step response framework, May 14, 2026.
- OpenAI: OpenAI Usage Policies — canonical policies governing how ChatGPT may be used.
- Anthropic: Core Views on AI Safety — Anthropic's public framing of the open research problems around LLM accuracy.
- Stanford HAI: Stanford HAI Research — ongoing research on the gap between LLM fluency and accuracy.
- Google Search Central: Search Essentials — the canonical Google documentation on what Google rewards.
- Search Engine Land: Why good content still loses in Google Search — SERP-side companion analysis, May 13, 2026.
- Search Engine Land: ChatGPT citations reward ranking and precision over length — April 16, 2026 study on what AI search actually rewards.
- Search Engine Land: Why AI search is your new reputation risk and what to do about it — April 3, 2026 piece on AI-search reputation dynamics.
- MIT Technology Review: Artificial Intelligence coverage — ongoing coverage of the fluency-vs-accuracy gap in modern LLMs.