ChatGPT ranking algorithm: how it picks brands to mention
Short answer OpenAI publishes no formula. It says ChatGPT ranks search results by "multiple factors" meant to surface relevant, reliable information, and that placement is not guaranteed. Studies back four signals: a page that fits the exact question, many reviews, mentions on other sites such as Yelp, and recent pages. A "30-day window" or a sentiment filter is unproven.
Want help applying this to your business? Talk it through on a free 30-minute call.
Book a callHow does ChatGPT decide which businesses to mention?
It answers from two places: what the model learned in training, and what it finds in a live web search. For the search part, OpenAI publishes one line:
"ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information. Placement is not guaranteed." (OpenAI Help Center: Searching the web with ChatGPT)
There is no published list of those factors. Everything below comes from outside studies, so treat it as evidence, not a formula.
| Route | What it draws on | Who it favors | How fast it changes |
|---|---|---|---|
| Memory (training) | Text the model read before its cutoff date | The most written-about names | Only when a new model ships |
| Live search | Results from search partners such as Microsoft, listings such as Yelp, business websites | Whoever best fits the question, with reviews and other sites to back it | Whenever the pages it finds change |
ChatGPT searches "when your question would benefit from current information" (OpenAI), which covers most "who should I hire" questions. Search is the route a small business can move. An audit of an OpenAI model across the 100 largest US metros found that from memory alone, only 11% of the doctors it recommended were real clinicians in that city; with search, 64% were. For financial advisers, firms with over 100 employees made up 55% of its picks from memory and 13% with search (Ibrahim and Zaki, 2026).
Which ChatGPT ranking signals are real?
Four have evidence behind them. The popular extras don't.
| Signal | What you'll hear | What the evidence shows |
|---|---|---|
| Fit to the exact question | "Build topical authority" | Real. ChatGPT rewrites the question into searches using your city and saved memories (OpenAI). The page that states the exact service, area and terms gets found. |
| Reviews | "AI filters out brands with bad sentiment" | Count matters more than stars. No source shows a sentiment filter. |
| Mentions on other sites | "One magazine feature beats 100 directory listings" | Real, but for a local business the directories and review sites are the main source. |
| Freshness | "A 30-day priority window" and "citation velocity" | Mild. ChatGPT cites newer pages than Google does, but its cited pages average over two years old. No study found a 30-day window. |
| Posting often | "Blog weekly to stay visible" | No evidence that frequency counts. |
| Video | "AI answers are going video" | Rare in ChatGPT: 1 YouTube citation in 7,651 answers in one 2026 study. |
Does ChatGPT favor fresh content?
A little. Ahrefs checked 17 million AI citations in 2025: pages ChatGPT cited were 958 days old on average, against 1,416 for Google's regular results. Fresher, yet still over two and a half years old. Old pages that answer the question still get cited.
Language models do react to dates. In a lab test, seven models, including OpenAI's GPT-4o, moved passages up when given a newer fake publication date, some by as many as 95 places (Fang et al., 2025). That tempts sites to change dates without changing the page. Google lists it as a warning sign:
"Are you changing the date of pages to make them seem fresh when the content has not substantially changed?" (Google Search Central: Creating helpful, reliable, people-first content)
OpenAI tells users to "review when it was published or updated" before trusting a cited page. So keep the facts that change actually current: prices, hours, services, service area, staff. Then show the true date of the change.
Do reviews affect ChatGPT recommendations?
Yes, mostly how many, less how high. In the same audit, restaurants that OpenAI's models recommended had 3 to 5 times the typical review count, but ratings only about a tenth of a star higher (Ibrahim and Zaki, 2026).
Records still show through when ChatGPT searches. With search on, 1.7% of the advisory firms it named had SEC misconduct records, against 5.1% of all firms. From memory alone, it was 18.3%. The researchers saw searched answers lean toward providers with cleaner records and better official ratings.
Nothing from OpenAI describes a "sentiment score". What to do: keep new reviews coming on the sites ChatGPT reads, and answer bad ones. On Yelp, don't ask: "We actually discourage business owners from asking their customers to write reviews" (Yelp for Business Support). See asking for reviews within the rules.
Do press mentions and directories count?
Yes. ChatGPT leans on what others say about you more than on what you say. In a 2025 University of Toronto study of product-ranking questions asked in the US and Canada, 69% to 92% of a web-enabled GPT's sources were "earned" sites, meaning reviews and independent publishers. Google's were 32% to 54%, and GPT used almost no social sites (Chen et al., 2025).
For local questions, the earned sources are mostly listings. Yelp appeared in 80% of ChatGPT's local answers, while Google Business Profile was "hardly used by ChatGPT" (BrightLocal, 2026). Your own site still counts: business websites made up 42% of all local AI citations.
So "quality over quantity" is half right. A local news "best of" list helps. A complete Yelp, BBB or trade directory page is not low value for a plumber or dentist: it is often the source ChatGPT reads. Where ChatGPT finds local businesses gives the order to fix them.
Does posting often or making videos help?
Not for ChatGPT, on current evidence. No study shows that publishing frequency counts. Google is blunt about it for its own search:
"Are you adding a lot of new content or removing a lot of older content primarily because you believe it will help your search rankings overall by somehow making your site seem "fresh?" (No, it won't)" (Google Search Central: Creating helpful, reliable, people-first content)
What counts is that a page answering the buyer's exact question exists. See does a small business need a blog.
Video is a Google and Perplexity play. Analyze, an AI search analytics firm, read 22,295 answers to 460 business-to-business questions in 2026. ChatGPT cited YouTube in 1 of 7,651 answers; Google AI Mode in 7.5% and Perplexity in 9.2%. OpenAI's search help describes citations, images and maps, not video. Make videos for customers, and put the key facts in page text too.
ChatGPT algorithm updates: what changed in 2026
OpenAI's release notes list model and feature changes, not ranking updates. These touch local businesses:
| Date | Change (OpenAI release notes) | What it means for you |
|---|---|---|
| Feb 9, 2026 | Ads test starts for US users on Free and Go | "ChatGPT’s answers are generated independently from advertising." |
| Jun 17, 2026 | Pulse, which sent Pro users proactive daily updates, "is being sunset"; Pro access ran 14 more days | Plans built on "predictive search" aim at a feature that is gone |
| Jun 24, 2026 | Model update: "Shopping and local business queries should also feel more useful." | With several requirements, answers are more likely to "clearly explain why a recommendation is a good fit." State yours plainly. |
| Aug 10, 2026 | Restaurant reservations: OpenTable, Resy (US), Yelp (US and Canada) | Restaurants: be bookable through one of the three |
| Oct 7, 2026 | GPT-6 becomes the chat model | Lists can shift with a new model. Retest. |
| Nov 10, 2026 | Business Hub: claim your listing, see mentions and clicks | Local businesses can claim only an existing US listing, for now |
Answers also change between identical runs. Repeat ChatGPT result sets overlapped only 23% on average (BrightLocal, 2026). One missing mention is not an update. Measure with the 20-question test.
Example
Illustrative cases. Numbers are invented.
A Dallas plumber who blogs weekly (US dollars)
He pays a writer $400 a month for four general posts. ChatGPT still names three rivals for "tankless water heater repair Dallas". They have 150 to 300 Yelp reviews; his Yelp page is unclaimed with 6. Better use of the money: one page on tankless repair with prices and area, a claimed Yelp page, and steady Google reviews. Retest in a month.
A med spa offered "citation velocity" for $1,500 a month
The pitch: 30 press releases a month to win "the 30-day window". No source shows such a window, and wire releases are the spa's own words, not earned mentions. A complete Yelp page and a steady stream of new reviews have evidence behind them.
FAQ
Is ChatGPT's ranking algorithm public?
No. OpenAI says only that it uses "multiple factors" and that placement is not guaranteed. Any tool claiming to know the formula is guessing from tests.
Can I pay to rank higher in ChatGPT?
No. OpenAI: "Advertising doesn’t influence ChatGPT’s independent answers." Claiming your Business Hub listing "doesn’t guarantee increased visibility" either (OpenAI Help Center: Setting up your business on ChatGPT).
Does ChatGPT give everyone the same list?
No. With memory on, "ChatGPT may use relevant saved memories when rewriting a search query" (OpenAI), and location changes results. Test in a temporary chat: by default it uses no memory or custom instructions (OpenAI release notes, August 2026).
Does ChatGPT still use Reddit?
Sometimes. In the 2026 Analyze study, 4.2% of ChatGPT's answers cited Reddit. See Reddit for a local business.
Different for you?
- ChatGPT names rivals, not you: how to get ChatGPT to recommend your business.
- You want the listings in order: where ChatGPT finds local businesses.
- You want a baseline first: the 20-question test.
- Someone is selling "citation velocity" or a GEO retainer: is GEO worth paying for?
- Your old pages need a refresh: update, merge or delete old pages.
- You also care about Perplexity, Gemini or Claude: rank in Perplexity, Gemini and Claude.
- You'd rather have it done and tracked: SEO & GEO Agent.