Why ChatGPT mentions your competitors and not your business

This is the one nobody checks, and it is invisible to every conventional audit.

On 20 September 2026 we tested eleven AI and search crawlers against seven domains. Ten received HTTP 200. One — a major AI system's corpus crawler, the one that builds its training and retrieval index — received HTTP 429 with an empty body on every single origin-bound request, while a normal browser received 200 on the identical URL in the same second.

The block was not in robots.txt, not in .htaccess, and not in any site code. It sat at the CDN edge, above the server, and the hosting provider confirmed it as a platform-side defect. The site owner had no way of knowing: the pages returned 200 to every browser and every other bot, and the provider's own dashboard reported the crawler as allowed.

Two things follow. robots.txt permission is not access — a crawler can be welcomed in robots.txt and refused at the edge. And testing robots.txt alone gives a false pass, because static cached files are often served to a crawler that is refused on everything else. The only honest test is to request an HTML page using the crawler's own user agent and read the status code.

Assistants resolve entities before they recommend. If your business name, address and phone number disagree between your Google Business Profile, your website, your directory listings and your company filings, there is no single confident answer to "who is this", and a system with a low-confidence entity will reach for a higher-confidence one instead.

This is the most common cause and the most fixable. It is also the one that overlaps most with ordinary local SEO, which is why the work is not exotic: consistent details, a verified profile, structured data that describes the organisation and links to the places it is independently listed.

Trading names that differ from licensed or registered entity names are a frequent source of this, and in some markets — the UAE especially — it is close to universal.

In August 2026 we examined sixty records across thirty commercial queries and two engines, looking at what each engine actually cited. The finding was that the citation mechanism is predicted by query specificity, not query difficulty.

Specific queries tend to resolve to a vendor's own domain. Broad comparison queries — "best X company", "top X agencies" — resolve overwhelmingly to third-party listicles and roundups. For that second class, no amount of work on your own site changes the outcome, because your site is not the kind of source the engine is selecting.

So before investing in a page, it is worth establishing which class your question belongs to. Effort against a listicle-locked query is spent whether or not it works, and it will not work.

A page can be reachable, correctly attributed and on a winnable query, and still not be quoted, because there is nothing on it that survives being lifted out of context.

We looked at ninety days of AI referral data against page structure and position. Two findings held. Rank is not the gate — a page at position 70 took twenty-one referrals in ninety days while a near-identical page at position 69 took none. And structural markup is necessary but not sufficient: pages carrying the full feature set at higher impressions than the cited ones took no referrals at all.

That second finding matters because it cuts against how this is usually sold. Structure is worth building and it guarantees nothing. Anyone promising citations in exchange for schema is describing something they cannot control.

Test crawler access properly. Request an HTML page with the crawler's user agent and read the status code. Not robots.txt, not a settings panel — a real page.

Resolve the entity. Make the name, address and phone identical everywhere, verify the profile, and make sure the structured data on your site says who you are and links to where you are independently listed.

Classify the query. Establish whether the question you want to win resolves to vendor pages or to listicles, before building anything for it.

Then make the answer extractable. A self-contained, accurate sentence that is true standing alone is the unit that gets quoted. A sentence that needs the paragraph around it to be accurate is not a citation candidate, and compressing it into one anyway is how businesses end up with a machine-readable claim they cannot stand behind.

It is not a guarantee of being recommended. Nobody controls what an assistant names, and any agency saying otherwise is selling something it cannot deliver.

It is not a replacement for technical SEO, local search work, paid media or conversion work. It is the same disciplines applied with retrieval in mind.

And it is not a reason to rename ordinary SEO as "AI SEO". Most of what makes a business retrievable by an assistant is work that was worth doing anyway.