Why We Checked an AI-VisibilitySales Pitch Instead of Ignoring It

A cold email claimed our site was invisible to ChatGPT and offered a free fix for it. The stat behind the claim was borrowed from the wrong industry, and the fix they were selling was the wrong move - but the site was, in fact, invisible, and the reason turned out to be bugs we had shipped ourselves.

The pitch: "half of buyers ask ChatGPT before Googling"

The email's opening claim was that over half of pet product buyers now ask ChatGPT for recommendations before they Google anything, that ChatGPT names three to five competitors when asked, and that our business was not one of the names. The offer: four to five new pages, built for free, that would fix it.

Two reactions are available to a pitch like that. Believe it and pay. Or dismiss it as spam and delete it. We did neither. We checked.

What we actually found

Logged out of ChatGPT, we ran the two prompts a real buyer would type. Both runs named exactly five competitors. Our business appeared in neither.

So the core complaint was true. The site was absent from the answers a shopper would actually see.

The stat and the fix were both wrong

The "over half" number had no source behind it, not in the email itself. The closest real number we could find is a G2 survey from April 2026, of over a thousand B2B software buyers - not pet retail. It's a different industry, cited as if it applied to ours.

The proposed fix was worse. Building four or five pages, each targeting one AI-search query, is close to what Google's own AI-search guidance warns against directly - prioritizing "AEO/GEO hacks" over ordinary SEO, and building the kind of doorway pages its spam policy targets. Ahrefs ran a controlled experiment on 34 self-promotional pages built for exactly this purpose: even when an AI engine cited them, it skipped them 43% of the time, and most citations didn't last.

What was actually breaking the site

None of what was actually wrong showed up in the pitch. We found it ourselves, on our own site, and every one of the bugs was ours:

• Over a dozen collection pages all pointed their canonical tag at the homepage, folding them into a single page as far as any crawler was concerned.
• The entire collections index rendered client-side only - zero links existed in the HTML a non-JavaScript crawler receives.
• Most of the FAQ answers existed in the page's code but never rendered into the HTML at all.

We had built the site ourselves, months before the cold email arrived. Every one of those bugs was ours to begin with.

What actually moves AI citation

The audit that found these bugs also turned up what the current evidence says does and does not move an AI engine's citations - useful beyond this one pitch:

• Being indexed and crawlable is the foundation. Google states there are no additional requirements to appear in AI Overviews or AI Mode beyond that, and Ahrefs' study of 1.4 million prompts found 88% of the URLs ChatGPT cites come from its own search retrieval.
• Server-rendered HTML matters, because most AI crawlers don't run JavaScript. ChatGPT's fetcher, and the OAI-SearchBot, ClaudeBot and PerplexityBot crawlers, all read a page's initial HTML - not what a script renders afterward.
• Off-site mentions beat on-site tweaks. Branded web mentions are the strongest correlate of AI Overview visibility, per Ahrefs' study of 75,000 brands, and 43.8% of ChatGPT-cited pages are third-party "best of" or "top" lists rather than a brand's own page.
• llms.txt is cheap hygiene, not a lever. Ahrefs checked 137,000 domains with one and found 97% received zero requests against it.
• Structured data is also hygiene, not a lever. Ahrefs' difference-in-differences test across 1,885 pages found schema markup moved ChatGPT and AI Mode citation rates by roughly 2%, statistically indistinguishable from zero.

Verify before you accept or reject

Rejecting the pitch outright, or paying for it on the spot, both only required ordinary skepticism: why would a stranger build you free pages that route the resulting leads straight to their own contact forms? That decision doesn't need research.

Getting the precise picture did. Going and checking the pitch's claims against primary sources - the actual guidance, the actual study, our actual site - is what separated the one part of the pitch that was true from the two parts that weren't, and it's what surfaced the real problem sitting underneath both of them.

If a similar email lands in your inbox, the fix isn't to trust it or trash it. Run the same two or three prompts a buyer would type, logged out, and see what actually comes back. Then check your own site for the same three things that were breaking ours: whether your product or category pages canonicalize to themselves, whether your navigation renders without JavaScript, and whether your FAQ content is actually present in the HTML a crawler sees, not just in the code behind it.

What's a piece of criticism you dismissed because of who it came from - and it turned out to be right about you?