GEO-005·Issue 01·Content & Semantics·5 / 132
Why Your Brand Has No Name in ChatGPT
When ChatGPT lists four competitors and never your brand, the gap is usually an unverifiable entity, not thin content. Three tests and a fix order that holds.
Issue focusThe Entry Point Shift: How AI Search Rewires Global E-commerce Traffic
A DTC pet-tech brand sells a self-cleaning litter box into the US. The founder asks ChatGPT for one that “won’t jam with clumping clay.” Four brands come back; none is hers. She rephrases; three new names, same omission. The Amazon listing converts. The store out-publishes both competitors. The team’s instinct is to write twenty more posts.
Most “the AI never mentions us” problems are entity problems — the model has no record of your brand, or has one it cannot verify — so more pages before the entity is fixed buy nothing.
Two kinds of invisibility, two opposite fixes
Two failures hide behind one symptom. Give the model your brand name alone and ask what it is.
If it has never heard of you, or recites your About page without knowing you are a company, the entity is missing. If it knows what you sell but not the warranty or the fit, the entity exists and the content is missing.
| What you see | Actual cause | Fix first |
|---|---|---|
| You never appear in category answers | No off-site sources | Off-site corpus, not pages |
| You appear, but only in your own words | Self-description only | Third-party evidence |
| You appear, but specs go unanswered | No answer unit | Citable facts on the page |
| Prompts treat you as two brands | Naming split | One name, one entity |
Three of four rows are entity work. Content tickets get answered with pages, which touch only the third row — so strong blogs stay invisible.
What makes a model write your name
Three conditions must hold at once.
Co-occurrence. Your name has to keep appearing beside the category, use case, and buyer in public text. It is not checking a business license; it learns from writing who gets discussed in a category.
Corroboration. More than one source must say the same thing. Calling yourself category-leading is a claim; a trade profile, a retailer listing, an independent test are evidence. Self-description proves you want to be a brand, not that anyone agrees.
Consistency. Brand name, legal entity, marketplace handle, the crowdfunding name from 2021 — three spellings read as three subjects, the cheapest reading for a model. Mismatch is close to non-existence.
A common counter-example: an About page of founder story, no legal entity, no address, one paid off-site mention. The model can confirm only that a site sells litter boxes, so it names none of you — and may never read the page if JavaScript builds it (can AI render your store?).
Three tests you can finish in one afternoon
No tooling, no vendor call. Record date, prompt, and screenshot.
- Nomination test. Ask the category three ways — “best X for Y”, “which X brands should I consider”, “is X or Z more reliable” — three times each. Pass is stable membership, not first place.
- Source test. Ask “which sources did you use” and look for one you do not control. If every citation is your own site or storefront, you have an evidence problem, not a copy problem.
- Consistency test. Ask “what is this brand and what does it sell” using the English name, marketplace handle, and any legacy name. Different categories, or blanks, mean fix naming first.
Answers shift by account, region, and day; one screenshot is not a baseline. Rerun in two weeks. Method: building an AI visibility baseline.
Fix in order: entity, evidence, content
Order is not a preference: unify the entity, or every review lands on a name the model files as someone else.
| Order | Action | Owner | Pass condition |
|---|---|---|---|
| 1 | One English name; legal entity, address, founding year | Brand + legal | Site, marketplace, profiles match |
| 2 | Organization markup on About; sameAs to profiles you control | Tech + content | Validator passes, names match |
| 3 | Trade press, retailer pages, independent tests | Marketing + content | Three sources, same entity, category |
| 4 | Category pages, comparisons, answer units | Content | A sentence stands without its context |
| 5 | Retest on the same prompt set | Growth | Before-and-after on file |
Step 2 is where teams go wrong: treating an encyclopedia entry as entity repair. Most cross-border brands miss the notability bar and the draft gets deleted. Substitutes: a trade profile, a retailer brand page, a community company page — citable, if not encyclopedic. Does your brand entity exist explains how to judge one.
Example: three names, one entity
Example — fictional, showing only the diagnosis and repair order.
One litter box, three names: an English name on the storefront, that name plus “-US” on the marketplace, a Chinese name on the 2021 crowdfunding page. The nomination test returns nothing, three times. The source test shows every cited page is theirs. The consistency test has the model describing “-US” as a different seller that stocks accessories.
Three repairs: collapse to one name and 301 the old URL; rewrite About with legal entity, warehouse address, founding year, and markup; give two flagship SKUs to independent testers.
Four weeks on, consistency passes — “-US” is no longer a second company. The nomination test surfaces the brand once, unstably. Not a win, but explainable: the entity moved from “mismatched” to “matched, not yet discussed.” Content compounds only from there.
Where this breaks
Skip it when the category has no nomination scenario: industrial buyers ask tolerances, MOQ, and lead time, not brands. Write specifications.
Skip it when marketplace volume dominates and the in-platform assistant reads only listings and reviews; repair attributes and stock first.
Skip it when the brand name is a generic word colonized by non-brand content. Renaming rarely pays, so standardize a “category term + brand” construction.
Entity repair is no substitute for fulfillment: vague delivery windows send reviews into the corpus. Without a native English editor, skip the third-party copy — a wrong description gets quoted back as fact.
More on this line in the content and semantics hub. Read how AI search split e-commerce traffic into four entry points next; entity work differs by entry point.
This month: pick your lead category, run all three tests, archive the screenshots. Then make the brand spelling identical on the site, the marketplace storefront, and the About page — capitalization, hyphenation, legal entity included. No budget, no API, and it decides whether the next hundred pages have a name the model can write down.