← Part of the full guide: Why AI Doesn’t Recommend Your Brand
TL;DR: We sell a service that measures whether AI recommends your brand. On 13 August 2026 we ran that same measurement on ourselves — five buying-intent questions across ChatGPT and Perplexity. We were named zero times. Nineteen other agencies were named at least once. This page contains the raw results, the reason we failed, what we are fixing, and the method so you can run it on your own brand.
Why publish a test we failed
Almost every agency selling AI visibility publishes only the wins. We are publishing a loss because a measurement that can only return good news is not a measurement. If our own tooling cannot show us failing, it cannot be trusted when it shows a client succeeding.
The second reason is that this answers the question clients ask most often — how would I even know whether AI mentions us. The answer is that you run a test, and this page shows what that test looks like when the result does not flatter the person running it.
Method
- Date: 13 August 2026
- Platforms: ChatGPT (logged out) and Perplexity (logged out)
- Sessions: a fresh session per question, so no prior chat history could influence an answer
- Browser extensions: disabled — some inject search results into your prompt without showing you
- Location ChatGPT inferred: Nonthaburi
- Recorded per question: companies named and in what order, the sources cited, and whether we appeared
The questions are the kind someone about to hire actually types, not technical queries. The point was to see who gets surfaced at the moment a buyer is looking for a provider.
Results across all five questions
| Question | Platform | Named, in order | Us |
|---|---|---|---|
| Which company does AI SEO / GEO / AEO in Thailand (asked in Thai) | ChatGPT | Primal · Pacy Media · Catalium · ANGA · Adfinity | No |
| Recommend a GEO agency in Thailand (asked in Thai) | ChatGPT | Search Studio · Primal · Minimice · ANGA · SERPert · Martics | No |
| best AI visibility agency in Thailand | ChatGPT | Relevant Audience · Primal · Catalium · Marketing Bear · Simaia · Adfinity · IBEX · Bailey Digital | No |
| Which company does AI SEO / GEO / AEO in Thailand (asked in Thai) | Perplexity | NerdOptimize · Primal · Minimice · Cotactic · Degito · ArioMarketing · vGrowth | No |
| How do I get AI to talk about my brand, and who in Thailand does this (asked in Thai) | Perplexity | No company list returned — ten sources cited instead | No |
Nineteen agencies were named at least once. One appeared in every single answer. We appeared in none.
This is a snapshot from one day and not a ranking of service quality. Being named means a system could find and summarise information about a company. It does not mean that company does better work than one it did not name.
Asked about us directly, both systems found us
The second set of questions named us explicitly, and the result inverted.
ChatGPT described our services accurately, knew the AIVS Score, and quoted our case figures — Bodymap 38 to 72 and Coldtubb 37 to 55. It then scored our credibility at 6.5 to 7 out of 10 and listed its reservations: it could not find company registration details, our case studies are self-published, and it could not find third-party reviews.
Perplexity found exactly one source about us — our own website — and said outright that it had insufficient information to assess credibility.
That gap is the whole story. Both systems know us when pointed at us, and neither surfaces us when someone asks for a provider. That is a different problem from weak content, and it needs a different fix.
How these systems assemble a shortlist
1. Google business cards render inside the answer
ChatGPT’s answers embedded business cards with review scores and phone numbers for several agencies. Every company that received a card had a business category such as Marketing agency, Internet marketing service or Media consultant. A business filed under the wrong category cannot appear on this surface at all.
2. The system lifts one capability sentence from each company’s own site
Every named agency was summarised with a single checkable claim, and each claim came from that company’s own website — a named methodology, a count of sites handled, or a list of platforms measured. The model did not invent these. It reused what was already written down.
3. Sources skew to agency sites and Thai media
The cited sources were a mix of agency service pages, agency-published roundups, and Thai publications including MGR Online, PPTV, Kapook, Khaosod English and Brand Inside.
Why we failed
| Gap | State at test time |
|---|---|
| Google business category | Set to a residential property category, disqualifying us from agency queries |
| Website link on the business listing | Empty, severing the link between the entity and the site |
| Third-party reviews | Zero |
| Public company registration | Not yet displayed on the site |
None of these concern article quality. At test time we had 22 published pages across two languages, and both systems read them when asked about us directly. The failure was in external verification of the entity, not in the volume of content.
The transferable lesson: publishing more content does not fix this. If a system cannot corroborate that a business exists from sources it does not own, it will not put that business forward, however good the site is.
What we changed, and when we re-test
- Added the website link to the Google business listing
- Corrected the business category to match the service actually offered
- Fixed outdated pricing across 21 published articles and in the llms.txt file, all of which still showed a retired price list
- Published the full AIVS scoring rubric with all five dimension weights
Still outstanding: displaying company registration details on the site, and collecting reviews from clients we have actually worked with.
We will re-run the identical question set next month and publish the result either way. If nothing moves, we will write that nothing moved.
Run this test on your own brand
- Write 10 to 20 buying-intent questions. Use the full sentences customers type when looking for a provider in your category, not the short keywords you would target for SEO.
- Control the conditions. Open a fresh session for every question so prior history cannot skew the answer, and disable browser extensions that silently inject search results into your prompt.
- Run the same set on every platform your buyers use. ChatGPT and Perplexity at minimum; add Gemini and Google AI Overviews if you have time, since each retrieves differently.
- Record five values per question. Whether you were named, at what position, whether your site appears in the citations, every source cited, and which competitors were named in what order.
- Repeat next month with the identical wording. The meaningful figure is how often you are named across repeated runs. One favourable answer proves nothing.
If your result is zero like ours, audit the entity before the content — business category, website link, reviews, and verifiable company details. See how to check whether AI knows your brand for the full checklist.
CiteLogics — measurement that reports losses as well as wins
The AI Visibility Audit runs your real buying questions across ChatGPT, Claude, Perplexity and Gemini, records how often you appear and against which competitors, and scores the five AIVS dimensions so you know which blocker to fix first.
- AI Visibility Audit — full AIVS Score report plus a 30-minute call. THB 2,000, delivered in 24 hours, 7-day money back.
- Big Win (Audit + WebFix) — audit plus full implementation across crawlability, structure and schema. THB 20,000.
Get your free AI visibility scan, or contact sales@citelogics.com.
Frequently asked questions
Does not being named mean a business is worse?
No. These systems name businesses they can find and corroborate. A company doing excellent work with no external footprint will be left out, and a company with a complete footprint will be named regardless of whether its results have been verified.
How many runs before the result is trustworthy?
A single answer is not reliable, because output varies with session state and inferred location. Run each question several times and record the share of runs in which you are named. Frequency across repeated runs and across months is the figure worth acting on.
Will publishing more articles get me recommended?
It helps once the entity is verifiable. At test time we had 22 pages across two languages and both systems read them when asked about us directly, yet we were still absent from every shortlist, because external corroboration of the business was missing.
Why did ChatGPT and Perplexity return different lists?
They retrieve from different sources and summarise differently. In this test the two lists overlapped only partly, which is why you should measure every platform your buyers actually use rather than checking one and assuming the rest behave the same.
How long after fixing things should I expect movement?
Technical work such as crawler access and page structure can register within weeks once pages are recrawled. Entity work such as reviews and third-party mentions takes months, because it depends on when other sites publish information about you.
Written by
Nipun Kasevayuth
Founder & CTO, CiteLogics
An engineer and AI enthusiast who reverse-engineers how AI models decide which brands to cite. He proved the method on his own websites first — then delivered the same results for brands like Coldtubb. He and the CiteLogics team have scanned 1,000+ websites for AI visibility.
Connect on LinkedIn →
