Brands rarely get excluded from ChatGPT answers on purpose. They get excluded because the evidence about them is thin, contradictory, stale or suspicious, and a model asked to recommend something it can defend picks the brands it can describe with confidence. Below are the eleven patterns we find most often in audits, in rough order of how much damage they do, each with the correction. If you fix the first three alone you will have done more than most of your competitors.
1. Your brand is a different entity on every platform
The most common and most expensive mistake. The website says “Acme, an AI-powered growth platform for ecommerce”. LinkedIn says “Acme Analytics, marketing software”. G2 lists you under “Business Intelligence”. A 2021 press release calls you “Acme Labs”. A directory has the old address.
To a model, those are four partially overlapping entities. When a buyer asks for “best analytics tools for Shopify stores”, the model cannot be sure Acme belongs, so it names the three competitors whose descriptions agree everywhere.
Fix: one legal name, one trading name, one category phrase, one two-sentence description, one address, one founding year, applied everywhere you control and requested everywhere you do not. The full approach is in brand entity consistency. For our done-for-you service we target first movement within the first week; how we do it is explained on your call.
2. Marketing copy where facts should be
Models quote plain statements. They hedge around slogans. If your About page says “We reimagine how teams collaborate” and nowhere states “Acme is project management software for agencies with 10 to 200 staff, founded in 2017, headquartered in Austin, priced from $12 per user per month”, there is nothing for a retrieval system to lift into an answer.
Fix: add a plainly written facts layer to your own site. What you do, for whom, where, since when, at what price, with which integrations or credentials. Mark it up with structured data. Keep the slogans; just stop making them carry the whole load.
3. Nobody else describes you in category terms
A brand that only describes itself is a brand with one witness. The research is consistent on this: in a University of Toronto audit of 1,516 queries, earned media accounted for 57% of GPT-4o citations (research roundup). The model is looking for independent sources that place you in the category the buyer named.
Fix: earn coverage that says the category word next to your name. A trade publication’s comparison, a podcast’s show notes, a partner’s integration page, an analyst’s list. Press releases that only you distribute do not count as independent.
4. Your coverage is from 2022
Recency weighs heavily. The Toronto audit found cited content had a median age of 62 days for Claude versus 130 days for Google in consumer electronics (same roundup). A Digital Authority Partners study found that only 10.6% of cited URLs persisted across 28 days, and 40 to 60% of cited sources rotate monthly (same roundup). A brand whose last substantial coverage is three years old is relying on sources that have mostly left the retrieval set.
Fix: treat coverage as a flow, not a stock. Something new, independent and factual about your brand should be appearing every month. Product updates, data you publish, customer stories told by the customer, expert commentary in the press.
5. You bought mentions, reviews or “AI visibility placements”
The temptation is understandable: if the model reads the web, put more of you on the web. But purchased mentions on low-quality roundup sites, seeded forum threads and bought reviews share three problems. They land on domains the model tends to discount. They describe you inconsistently, because the seller wrote them. And they sit next to the “is this brand legit” question that buyers also ask, where a sudden burst of thin praise reads as exactly what it is.
In our experience these tactics underperform, and we do not use or recommend them. The comparison is laid out in buying brand mentions versus earning the recommendation.
Fix: redirect the budget to earned coverage, real review generation on platforms your buyers use, and the facts layer from point 2.
6. You assumed Google rankings would carry over
Years of SEO equity feel like they should count. The overlap data says otherwise: AI-cited sources matched Google’s top 10 only 4.0% of the time for GPT-4o, 11.1% for Gemini, 12.6% for Claude and 15.2% for Perplexity in the Toronto audit (same roundup). The pages ChatGPT reads for a buyer question are mostly not the pages Google ranks for it.
Fix: keep the SEO; it still drives the brand search buyers do after an AI recommendation. But audit ChatGPT separately and work on its source set directly. The differences and overlaps are covered in our generative engine optimization guide.
7. You are absent from the review platforms the model actually reads
Reviews matter twice: as a trust signal (78% of consumers in a 2026 Idea Grove survey say reviews raise trust (citybiz)) and as a source the model retrieves for category and reputation prompts. The mistake is not having zero reviews; it is having them on the wrong platform. A B2B SaaS brand with 400 Google reviews and no G2 or Capterra profile is invisible on the pages ChatGPT pulls for “best [software] for [use case]”.
Fix: find out which review domains are cited for your category prompts (your audit will show this), build a presence there, and keep it current.
8. You have multiple brands, domains or sub-brands for the same thing
Common in ecommerce, finance and restricted categories: a parent brand, two product brands, a regional domain, an old brand still live. Each gets a share of the coverage and none gets enough to be named with confidence. Worse, the model may merge them with each other or with unrelated companies.
Fix: consolidate where possible. Where you cannot, make the relationships explicit on every property (“Acme Pay is a product of Acme Inc.”) so the model can attribute coverage to one entity.
9. You never check the “is it legit” answer
Teams measure category prompts and ignore the brand-name prompts. Then a buyer asks “is Acme safe?” and the answer is “I could not find much information about Acme”, or worse, a hedge anchored to one old complaint thread. The category mention got you shortlisted; the reputation answer took you off.
Fix: audit the trust prompts alongside the category prompts and fix what they expose. We cover this in detail in what ChatGPT says when buyers ask if a brand is legit.
10. You tested once, in your own logged-in account, and drew conclusions
Someone on the team types a prompt into ChatGPT, sees a competitor, and the conclusion “ChatGPT hates us” or “we’re fine” becomes company fact. But answers vary by session, by logged-in state (which carries your history), by device, by country and by day. A single check is an anecdote.
Fix: sample. Our sampling method uses about 100 real answers per prompt per market from genuine mobile connections in the target country, each classified for brand named, position, link present and competitors. That produces a confidence band. If you are doing it yourself, the ladder is: 3 runs per prompt shows a pattern (the free tool), 5 to 10 runs is a DIY baseline, 20 runs is the minimum for a figure you would act on, and about 100 answers per prompt is what a full audit uses for a confidence band. Run from the target country in fresh sessions.
11. You optimized for the wrong prompts
The final mistake is doing everything right for prompts buyers do not use. Marketing teams write prompts the way they write ad copy: “best enterprise-grade AI growth platform”. Buyers write “software to stop losing track of client emails, small agency, under $50 a month”. The second one is a prompt cluster you can actually be named in.
Fix: build prompts from how buyers talk, not how you talk. Our ChatGPT prompt generator tool produces five buyer-style prompts for any category, location and buyer type as a starting point, and the prompt clusters guide explains how to expand that into a measurable cluster.
The pattern behind all eleven
| Mistake | What the model experiences | Correction in one line |
|---|---|---|
| Inconsistent entity | Four brands that might be one | One set of facts everywhere |
| Slogans not facts | Nothing quotable | Plain facts layer on your site |
| Only self-description | One witness | Earned coverage in category terms |
| Stale coverage | Sources that have rotated out | Monthly flow of new independent content |
| Bought mentions | Suspicious, low-weight sources | Redirect to earned and real reviews |
| SEO assumption | Different source set | Audit and work ChatGPT directly |
| Wrong review platforms | Absent from retrieved pages | Be where the category is cited |
| Fragmented brands | Diluted or merged entity | Consolidate or make relationships explicit |
| Unchecked trust prompt | Hedged or thin verdict | Audit and fix reputation prompts |
| One anecdotal test | Noise mistaken for signal | Sample at scale |
| Wrong prompts | Optimized for queries nobody types | Build from buyer language |
Every row is a version of the same thing: the model recommends what it can describe confidently from consistent, recent, independent evidence. Make that true of your brand and the mention rate follows. For the full explanation of why competitors get named when you do not, see the seven reasons ChatGPT skips a brand.
If you would rather have the audit and the fixes handled, our done-for-you service does both, with prompt-level reporting you can reproduce in ChatGPT and a monthly re-audit, no lock-in. Targets, not guarantees; how we do it is explained on your call. Start by seeing what ChatGPT says about your brand at the Blue Ocean GPT homepage.
What to do next
- Audit your own entity across site, LinkedIn, review profiles and directories this week and reconcile every difference.
- Write the plain facts page your site is missing and add structured data.
- Sample your top 20 buyer prompts properly before deciding which of the other eight mistakes applies to you.
Frequently asked questions
What is the single biggest GEO mistake?
Inconsistent brand facts across the web. If your name, category, location or description differ between your site, LinkedIn, review profiles and directories, the model has to guess which version is true, and it hedges or omits. Fixing entity consistency is cheap, fast and usually produces the earliest measurable movement in mention rate.
Does buying brand mentions help ChatGPT visibility?
In our experience it underperforms and carries risk. Purchased mentions tend to land on low-quality pages the model discounts, they rarely describe your brand in consistent category terms, and they do nothing for the trust prompts buyers also ask. Earned coverage in factual context is what the research associates with citations, and it is what moves mention rate in practice.
Can good Google rankings hurt my ChatGPT visibility?
Not directly, but relying on them can. A University of Toronto audit found only 4.0% overlap between GPT-4o's cited sources and Google's top 10. Brands that assume SEO will carry over often stop there and never build the third-party, recent, consistent footprint that AI answers draw on. The mistake is the assumption, not the rankings.
How do I know if a mistake on this list is costing me mentions?
Measure. Sample your buyer prompts many times, record whether you are named and which sources are cited, and compare with the brands that are named. The pattern of cited sources usually points directly at the gap: a competitor appears because of a review platform you are absent from, a trade publication that never covered you, or a plainly written fact page you do not have.
Are these mistakes the same for Gemini and Perplexity?
Largely yes. All three engines reward consistent entities, recent independent coverage and quotable facts. Perplexity and Gemini cite a somewhat different source mix, and the overlap of cited domains between platforms is only 2 to 17% according to a Digital Authority Partners study, so the specific pages differ. The underlying habits that make a brand legible to one tend to help with all of them.
Find out what ChatGPT says about your brand before you spend anything.
Apply in about two minutes. We check your keywords, then run your real buyer prompts live on a 30-minute call. If we cannot create the result for your keywords, we tell you on the call, not after an invoice.
See if my keywords qualify →