Why does asking ChatGPT about your own brand prove nothing?
Because the name in the question does the work. Ask “Is Hollow Oak Bakery any good?” and ChatGPT searches for Hollow Oak Bakery, finds your website and reads it back to you. A buyer who does not know you asks “best sourdough bakery in Boise”, and that question has to be answered from what other people have written. Only the second question brings in customers.
Your own account skews it further. OpenAI’s help page says that “if memory is enabled, ChatGPT may use relevant saved memories when rewriting a search query,” and that ChatGPT “may use an approximate location based on your IP address to provide relevant local results.”[1] Months of chats about your business are a memory of your business. The same page explains that ChatGPT search “sometimes partners with other search providers” and “typically rewrites your query into one or more targeted queries that it sends those providers,” so the answer depends on what those providers have indexed, not on the model alone.[1]
Then there is variance. When Pluspoint put the same dining questions to ChatGPT twice in August 2026, identical questions returned identical sources 8 times in 100, and the two source lists overlapped by 38% on average.[8] One answer is a sample of one.
How do you run the five-question audit?
Five unbranded buyer questions, a clean chat with Search on, three runs over two days, and a note of who was named and which websites were used. About an hour in all.
- 1
Write five questions a buyer would ask
Use a stranger’s words, never your name. A dentist: “best dentist in Layton for a nervous adult”. A skincare brand: “best fragrance-free moisturizer for rosacea under $40”. Mix one “best X in [city]”, one “X for [specific need]”, one “alternatives to [big rival]” and two in your customers’ own words. - 2
Open a clean chat
Start a Temporary chat and choose Unpersonalized, which OpenAI says “does not use memory, custom instructions, or plugins.”[3] Or sign out. - 3
Turn Search on
Select View all tools, then Search, before each question; left alone, ChatGPT searches only when it decides “your question would benefit from current information.”[1] An answer with no citations came from the model’s memory, not the web. - 4
Ask which websites it used
Select Sources to “view cited sources and other relevant links,” then ask: “Which websites did you use to answer that?”[1] Write the domains down. They are the pages you need to be on. - 5
Repeat three times over two days
Run all five again later the same day and once more the next day. One run cannot tell a pattern from a coin flip when the same question so rarely returns the same sources twice.[8]
What does the scoring sheet look like?
One row per question, one column per run, and a last column for the sources. Score each run as you, a named rival, or nobody. A business named in three of three runs is a stable recommendation; one named in one of three is noise.
| Buyer question | Run 1 | Run 2 | Run 3 | Sources ChatGPT named |
|---|---|---|---|---|
| Best sourdough bakery in Boise | Rival A | Rival A | Nobody | Local newspaper “best of” list, Reddit thread, Yelp |
| Bakery that does gluten-free birthday cakes near downtown Boise | Rival B | You | Rival B | Tourism board page, two food blogs |
| Where to order bread for a 40-person office lunch in Boise | Nobody | Nobody | Nobody | No local pages; national catering chains |
| Alternatives to the grocery-store bakery in Boise | Rival A | Rival C | Rival A | City magazine, Reddit thread |
| Bakery in Boise open before 7 a.m. | You | You | You | Your hours page, a local news story |
Read it in three lines. Count the runs where you were named: here, four of fifteen. Count the runs where nobody was named: a question with no incumbent is the cheapest to win. Then tally the domains in the last column; those pages decide the answer, and most are pages you can be on. Keep the sheet; next month’s copy shows movement.
What decides whether ChatGPT names you?
Pages other people publish about you, more than anything on your own site. OpenAI’s description is short: ChatGPT “ranks search results using multiple factors intended to help users find relevant, reliable information. Placement is not guaranteed.”[1] Two studies fill in the detail.
Pluspoint’s August 2026 audit of 210 ChatGPT dining answers found city and national editorial supplied 48.9% of cited domains, small independent sites 18.6%, review and booking platforms 11.4%, forums 10.9%, and listing pages on Google, Bing, Apple or Yelp none.[8] SE Ranking’s November 2025 study of 129,000 domains found roughly four times the citation odds for domains with heavy Reddit and Quora mentions, 4.6 to 6.3 citations for domains on review platforms against 1.8 for those absent, and 6.0 citations for pages updated within three months against 3.6.[7]
Two things are in your hands. Being crawlable: sites opted out of OAI-SearchBot “will not be shown in ChatGPT search answers.”[4] And consistency: ChatGPT rewrites your question, sends it to search partners and builds an answer from what comes back, so a business whose name, hours and offering read the same everywhere gives the model one story to tell.[1] Paying is not a route: “advertisers have no ability to shape, rank, or alter ChatGPT’s responses.”[2]
Does the same check work for Gemini and Perplexity?
Yes, with the same five questions, and expect different answers. In Pluspoint’s seven-industry count, ChatGPT and Gemini cited the same sources for the same question 8% of the time and named the same top business 4.2% of the time.[9] Citing Steady Demand’s fifty-metro data, Pluspoint puts nearly 60% of Gemini’s local citations on the business’s own website, where ChatGPT leans on what others publish.[9]
Google’s rules are plain. Its page on AI features says: “You don’t need to create new machine readable files, AI text files, or markup to appear in these features. There’s also no special schema.org structured data that you need to add.”[5] To be a supporting link in AI Overviews or AI Mode, “a page must be indexed and eligible to be shown in Google Search with a snippet.”[5] Ordinary search eligibility is the whole ticket.
Perplexity shows its work most openly. Its help center says it gathers “information from authoritative sources like articles, websites, and journals” and that each answer “includes numbered citations linking to the original sources.”[6] In Profound’s analysis of 680 million citations from August 2024 to June 2025, Reddit was Perplexity’s most cited domain at 6.6%; Wikipedia led on ChatGPT at 7.8%.[10] Score each engine on its own sheet. Showfolk’s AI search tracker runs the three side by side every week.
How often should you repeat it, and what should you change first?
Monthly by hand, weekly if a tool does it for you, and always the same five questions so the sheets compare. One sheet says where you stand; the next says whether anything moved. SE Ranking frames the update cadence as every two to three months, so a monthly check catches a change inside one cycle.[7]
That last step is the one owners skip. If every answer names the rival for its Sunday hours or its gluten-free range, an ad that answers the same question meets the same buyers before they ask an assistant. Showfolk runs this check every week across ChatGPT, Gemini and Perplexity, records who was named and why, and writes the ad that answers it: how the AI search tracker works. For using ChatGPT itself to draft those ads, see Can ChatGPT make Facebook ads?
Questions people ask
- Does ChatGPT recommend my business?
- Ask the questions a buyer asks, without your name, in a temporary unpersonalized chat with Search on, three times over two days. Named in most runs means yes. If a rival is named, open Sources to see which pages decided it; those are the pages you need to be on.
- How does ChatGPT choose which businesses to recommend?
- OpenAI says ChatGPT ranks search results on multiple factors, that placement is not guaranteed, and that a site must allow OAI-SearchBot to be eligible. Audits of its citations find it leans on editorial pages, forums and review platforms rather than business listing pages.
- Can I pay to be recommended by ChatGPT?
- No. OpenAI’s help page on ads says ads do not influence ChatGPT’s answers and that advertisers have no ability to shape, rank or alter responses. Ads are labeled as sponsored and appear below an answer, separate from it.
- Why does ChatGPT give a different answer each time I ask?
- Because each run is a fresh web search, often through search partners, and the model writes a new answer from whatever comes back. In an August 2026 audit, identical questions returned identical sources 8 times in 100. Three runs over two days separate a pattern from a one-off.
- Do I need special schema or an llms.txt file to show up in AI answers?
- Not for Google. Its AI features page says no new files, markup or special schema.org data are needed; a page has to be indexed and eligible for a snippet. SE Ranking’s ChatGPT study likewise advises skipping llms.txt.
- How often should I check whether ChatGPT recommends my business?
- Monthly by hand, or weekly with a tool, always with the same questions so results compare. A monthly check is frequent enough to see whether a change to your site or your coverage moved the answer.
Sources
Every figure above points at one of these. Facts were last checked on .
- [1]OpenAI Help Center: ChatGPT search. How to turn Search on, the Sources panel, search partners and query rewriting, memory and location in queries, and “Placement is not guaranteed”
- [2]OpenAI Help Center: Ads in ChatGPT. “Ads do not influence ChatGPT’s answers” and “advertisers have no ability to shape, rank, or alter ChatGPT’s responses”
- [3]OpenAI Help Center: Temporary chat in ChatGPT. The Unpersonalized option “does not use memory, custom instructions, or plugins”
- [4]OpenAI developer docs: Overview of OpenAI crawlers. What OAI-SearchBot does; sites opted out “will not be shown in ChatGPT search answers”
- [5]Google Search Central: AI features and your website. No special files, markup or schema needed; a page must be indexed and eligible for a snippet
- [6]Perplexity Help Center: How does Perplexity work?. Sources it gathers from, and numbered citations on every answer
- [7]SE Ranking, November 24, 2025: How to optimize for ChatGPT (study of 129,000 domains and 216,524 pages). Reddit and Quora mentions (roughly 4x), review platforms (4.6 to 6.3 vs 1.8 citations), freshness (6.0 vs 3.6), and the advice to skip llms.txt
- [8]Pluspoint, August 31, 2026: How to get ChatGPT to recommend your business. 210 ChatGPT answers to dining questions across eleven US cities; the source-type table, zero listing pages, and 8 in 100 identical source sets
- [9]Pluspoint, September 1, 2026: Who decides which local businesses AI recommends. ChatGPT and Gemini agreement rates (8% sources, 4.2% top business) and Gemini’s reliance on the business’s own site, citing Steady Demand
- [10]Profound: AI platform citation patterns (680 million citations, August 2024 to June 2025). Reddit 6.6% of Perplexity citations; Wikipedia 7.8% of ChatGPT citations
