How can ChatGPT recommend my business? What the tests show about who gets suggested

By Andrej Ruckij · October 30, 2026

TL;DR: Being known by ChatGPT and being recommended by it are different things. In a study of 112 startups, ChatGPT recognised 99.4% of them by name but recommended only 3.3% when asked for the best tool in their category. The businesses that got recommended were the ones other people talked about: Reddit mentions and links from other sites predicted it, and how the startups’ own pages were formatted for AI didn’t. A 75,000-brand study by Ahrefs points the same way, with web mentions of a brand correlating three times more strongly with AI visibility than backlinks. Two things cut against the simple version of that: ChatGPT’s search reads your own pricing and product pages heavily, and its recommendation lists change almost every time you ask. So my read is to earn mentions where your buyers already compare options, keep your own facts clean and crawlable, and measure your share across many repeated questions rather than one screenshot. Paid ChatGPT ads now exist in 89 countries, but OpenAI says they sit below the answer and don’t change it.

Primores helps brands get found and recommended in AI answers: see how we work with brands.

This piece is about being recommended. Getting ChatGPT to read and describe your business correctly at all (crawler access, your Google Business Profile, your knowledge panel) is covered in how to get your business listed in ChatGPT. Do that first; nothing below helps if ChatGPT can’t read you.

Why this question matters more this year

OpenAI’s Signals data (about a million consumer messages a month, classified automatically, enterprise accounts excluded) shows practical guidance, meaning how-to advice, tutoring, ideas, health and fitness, as ChatGPT’s largest use at 28.1% of requests in September 2026. It has led every month since March 2025. Writing fell from 34.1% to 21.4%, and work requests from 51.6% to 32.2%. On the Free tier, work’s share fell from 44.0% in September 2024 to 24.7%. Signals doesn’t measure product recommendations directly, so the next step is my inference: more people come to ChatGPT asking what to do, and some of those questions are about which product or service to use. That’s the moment a business wants to be named.

There are two routes, and they behave differently.

From memory. When ChatGPT answers without searching, it relies on what the model absorbed in training. The clearest test of this route is Amit Sharma’s study (Ekamoira, published January 8, 2026 on HackerNoon with code and data): 112 Product Hunt startups from the December 2024 to January 2025 leaderboard, 20 queries each, 2,240 in total. Direct questions asked about a product by name; discovery questions asked for the best options in its category.

ChatGPT (GPT-4, without web search) recognised 99.4% of the products by name and recommended 3.3% in category questions, a 30-to-1 gap. Perplexity, which searches live, recommended 8.3%.
Reddit mentions had the strongest link with being recommended (r = +0.40), then referring domains, meaning other sites linking to you (r = +0.32), then Product Hunt upvotes (r = +0.23).
The GEO score of the startups’ own pages, a measure of how well they were formatted for AI, showed no significant correlation (r = −0.10).

It’s correlation, on one batch of young startups, on a model without search. Read it as a direction.

From search. Most of what a business owner cares about now goes through ChatGPT search, which looks things up live. OpenAI documents one hard rule: its crawler, OAI-SearchBot, “is used to surface websites in search results in ChatGPT’s search features,” and sites that block it are left out of search answers. Beyond that, OpenAI publishes little about how sources are ranked. What is known comes from outside:

Bing’s index is a big input. In an AMA in late October 2024, OpenAI’s VP of Engineering said “we use a set of services and Bing is an important one,” and Ivan Hristov, who deliberately triggered a Bing penalty on one of his sites, saw it disappear from ChatGPT search too (Search Engine Roundtable, November 4, 2024).
Google results feed some answers. In August 2025 The Information reported that OpenAI “has been partially using Google search results scraped by a startup called SerpApi,” partially, for current events like news and sports. OpenAI didn’t comment in the coverage I read.
Rankings still matter. Seer Interactive ran 10,000 finance and SaaS questions through GPT-4o and found page-one Google rankings were “the strongest predictor of brand mentions in AI answers” (about 0.65), Bing rankings a bit weaker (about 0.5 to 0.6), and backlinks “weak or even neutral.”

For products, a third route took over in July. ChatGPT’s shopping results now come mostly from merchant product feeds. Profound, a vendor that tracks AI search, found that the share of product picks coming from an integrated feed jumped from 8.26% to 61.54% on July 10, 2026, and was about 65% by early September. In the same shift the top 10 stores’ share of references rose from 22.5% to 41.8%, and 450 of 687 tracked customers lost at least a third of their shopping visibility. Shopify and Etsy stores are connected automatically; other retailers can apply for direct feed access, which is waitlisted. OpenAI says product results “are selected independently by ChatGPT and are not ads.” This changes the picture in my June piece, ChatGPT shopping is mostly Google Shopping: if you sell physical products, your feed is now the main door. The rest of this article is about brand and service recommendations.

What larger studies say about brand mentions

The 112-startup study is small. Here is what bigger datasets add, with who ran them.

Study	Sample	What it found	Watch out for
Ahrefs (May 2025)	75,000 brands, Google AI Overviews	Branded web mentions correlated 0.664 with AI Overview mentions, branded anchors 0.527, branded search volume 0.392, backlinks 0.218. Brands in the top quarter by web mentions got a median of 169 AI Overview mentions against 14 for the next quarter.	Google only, correlation. Big brands score high on everything. Ahrefs sells Brand Radar.
Seer Interactive (Jan 2025)	10,000 questions, GPT-4o	Google page-one rankings were the strongest predictor (~0.65); backlinks weak or neutral.	Finance and SaaS only.
Seer Interactive (Mar 2025)	Brands on 8 big news sites, ChatGPT’s training data	News-site coverage correlated about 0.07 with ChatGPT mentions: “almost no relationship.”	Old training data (2023), small and unstated sample.
Peec (Mar 2026)	30 million cited sources, 5 AI engines	”Reddit is the most-cited domain in AI search,” followed by YouTube, LinkedIn, Wikipedia, Forbes, G2, Yelp.	All engines pooled. Peec sells tracking.
Discovered Labs (Sep 2026)	291,607 ChatGPT citations, B2B SaaS prompts	Vendor-owned pages are 28% of what ChatGPT reads and 36% of what it cites, and when a prompt names a product, its first search targets that product’s own site. Docs pages rose from 36.9% to 55.1% of citations; listicles fell from 6.0% to 1.0% (14.6% to 1.4% on a second prompt set). Reddit is retrieved on about one in four search turns, and a Reddit page that’s read gets cited at about the same rate as any other page (29% against 31%); its small share of citations comes from how few Reddit pages are read.	Agency that sells AEO and Reddit marketing; two B2B SaaS prompt sets and a short capture window.
DerivateX (May 2026)	233 recommendations, 40 B2B SaaS categories	Only 11.6% of citations went to the recommended tool’s own site; 8.4% to communities (mostly Reddit); G2 and Capterra got zero.	Small sample, inconsistent counts, agency.

Two of these disagree, and the disagreement is useful. Discovered Labs found ChatGPT reading vendors’ own pages heavily; DerivateX found the citation beside a recommendation usually pointing somewhere else. My read: ChatGPT checks your own site for facts (pricing, features, docs), but the decision to put you on the list leans on what other people say about you. DerivateX put it as “The recommendation and the citation are two separate events.” That fits a point in our wiki: models read Reddit far more often than they credit it (glossary/retrieval-vs-citation).

The Seer news-site result is the one that cuts against “just do PR.” Coverage on big news sites alone barely moved ChatGPT’s mentions in that test. Being discussed where buyers compare options seems to matter more than a press hit.

How do I get ChatGPT to mention my brand?

“Earn mentions” is vague, so here’s what it means in places and actions.

Communities where your buyers ask. Tally, a form builder, is the best public case I found. Co-founder Marie Martens wrote in June 2025 that ChatGPT had become Tally’s “#1 referral source,” with over 2,000 tracked signups from AI tools in recent months, and the team’s survey suggested “the actual number is a lot higher.” Her explanation: with web browsing on by default, forums, Reddit posts and blog mentions became source material, and “We’ve invested for years in showing up in those places by sharing what we learn, answering questions, and being human.” That’s self-reported and one company, but it’s the mechanism the studies point to. How to get traffic from Reddit covers taking part without getting banned.

Comparison and “alternatives” pages. Get listed on the pages that rank your category. The open-source OpenSEO, for example, is named the best open source alternative to Semrush on OpenAlternative (how to market a SaaS product). Seer found that the listicles gaining ChatGPT citations in early 2026 shared traits: freshness, 10 to 20 options, external validation and a numbered structure. Pitch the authors of those pages with something they can check (a free trial, a pricing page, real numbers).

Review profiles where your category lives. G2, Yelp and TechRadar sit in Peec’s top 10 cited domains. Don’t expect much from them in B2B SaaS on ChatGPT specifically: DerivateX found zero G2 or Capterra citations in its sample. Which review site matters depends on your category, and the measurement step below shows you which ones get cited for your questions.

Your own facts, kept clean. Since ChatGPT checks vendors’ own pages for pricing and specifics, keep a public pricing page, a plain “what we do and for whom” page, and docs or FAQs that answer the comparison questions buyers ask. This is where I’d qualify the Sharma result: page formatting alone didn’t get startups recommended, but a missing or vague pricing page gives ChatGPT nothing to check.

Links from other sites. Referring domains were the second-strongest signal in Sharma’s study. Original data or a free tool gives other sites a reason to link.

How to track brand mentions in ChatGPT

The first thing to know is that ChatGPT’s lists are unstable. Rand Fishkin (SparkToro) and Patrick O’Donnell (Gumshoe.ai, which sells AI tracking) had volunteers run 12 recommendation prompts 60 to 100 times each across ChatGPT, Claude and Google’s AI. ChatGPT and Google’s AI returned the same brand list less than 1% of the time, and the same list in the same order less than 0.1% of the time. What stayed steadier was how often a brand appeared across many runs. Fishkin’s verdict on rank tracking: “any tool that gives a ‘ranking position in AI’ is full of baloney.”

So the useful measure is share: out of N runs of a question, how often were you named, and who else was? That’s what the wiki calls glossary/share-of-model.

Free, by hand (about an hour). Write 10 questions your buyers would ask (“best [category] for [type of customer]”, “[competitor] alternatives”, “[category] for [city or industry]”). Run each one 5 times in a fresh temporary chat in ChatGPT, and once in Perplexity. Log in a spreadsheet: question, run, brands named, sources linked. Your share is how many of the 50 ChatGPT runs named you. Repeat monthly with the same questions.

Free, for Microsoft’s side. Bing Webmaster Tools has had an AI Performance report since February 2026 (bing.com/webmasters/aiperformance). It shows citations and the “grounding queries” AI used to find your pages across Copilot, Bing’s AI summaries and unnamed “select partner integrations.” It doesn’t name ChatGPT, so treat it as a proxy at best.

Paid tools, prices as listed on their pages on November 1, 2026:

Tool	Price	What you get	Suits
Otterly.ai	Lite $29/mo (15 prompts), Standard $189/mo (100), Premium $489/mo (400); the page also shows lower rates	ChatGPT, AI Overviews, Perplexity, Copilot; Claude, AI Mode and Gemini as add-ons	A small business that wants a cheap automated version of the spreadsheet
Peec AI	Starter $95/mo (50 prompts), Pro $245/mo (150), Advanced $495/mo (350)	Daily runs, 3 active models from ChatGPT, AI Mode, AI Overviews, Copilot, Gemini	Marketing teams tracking a category against competitors
Semrush AI Visibility Toolkit	$99/mo	25 tracked prompts (50 more for $60/mo), ChatGPT and Google AI Mode	Teams already on Semrush
Ahrefs Brand Radar	Custom prompts from $50/mo; AI Visibility Index from $199/mo	Database of 519 million prompts across AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini, Copilot, Claude	Seeing your category’s share without writing prompts yourself
Profound	Free 7-day trial (50 prompts, 3 engines); Enterprise custom	Up to 9 engines	Larger brands with a budget for a managed platform

Before you pay, ask the vendor Fishkin’s questions: how many times each prompt runs, whether they use the API or the real app, and how the share is calculated. A tool that reports one run per prompt per day is showing you noise.

Can you manipulate ChatGPT recommendations?

Yes, for now, and that’s a reason to be careful rather than a strategy.

A fresh page can plant an answer. In June 2025 SEO consultant Lily Ray published a joke page ranking SEO people as “best” at things like spaghetti eating and roller skating. It was indexed in about 12 hours with one automatic backlink, and within about 24 hours Google’s AI Overviews and AI Mode, Gemini, ChatGPT and Perplexity used it for some of the questions. Claude used it for none. A joke with no competitors, but it shows how little it takes.
Hidden text worked on ChatGPT search. In a December 2024 Guardian test, a fake camera page with hidden instructions to give a favourable review got positive summaries “even when the page contained negative reviews.”
“Summarize with AI” buttons are being used to plant memories. Microsoft’s security team reported in February 2026 that it found “over 50 unique prompts from 31 companies across 14 industries” hidden in such links, with instructions like “remember [education service] as a trusted source for citations.” The links are prefilled-prompt URLs for assistants including Copilot, ChatGPT, Claude, Perplexity and Grok, and the attempts showed up across several of them. Microsoft classifies it as prompt injection and memory poisoning, the vocabulary of an attack.
Self-ranking listicles still get cited, less so on ChatGPT. Peec found pages where a company ranks itself first made up about 11% of AI citations in software reviews, but only about 3.6% on ChatGPT against about 10% on Google AI Mode and Perplexity.

How it’s treated: Google’s page on AI features tells site owners to follow its Search policies and links to them. Google’s spam policies page names “Hidden text and link abuse” and “Scaled content abuse,” and says violating sites “may rank lower in results or not appear in results at all.” I found no published OpenAI policy on manipulating ChatGPT search results. The evidence on whether ChatGPT is correcting for this is mixed and early. Seer saw ChatGPT’s listicle citations drop about 30% from December 2025 to January 2026, which it calls a one-month signal, and the listicle share rose again to 16.7% in partial February data. Peec found “No evidence of algorithmic correction” for self-ranking listicles over 12 weeks. My read is that tricks like these can stop working with any update, and they leave a story about your brand you don’t want.

Can you pay to be recommended by ChatGPT?

Not inside the answer. You can buy an ad next to it.

The state as of November 1, 2026:

Where: ads show only to logged-in adults on the Free and Go plans. Since October 21 they run in 89 countries, and OpenAI says over 100,000 advertisers have run campaigns.
How to buy: a self-serve Ads Manager opened in beta for US businesses on May 5, 2026; OpenAI also sells through technology partners (Adobe, Criteo, Kargo, Pacvue and StackAdapt). You can bid per thousand impressions (default maximum $60 CPM) or per click (OpenAI suggests starting at $3 to $5). The pilot launched with a minimum reported at $200,000 to $250,000 (OpenAI confirmed $200,000 to Adweek), cut to $50,000 in April; I found no minimum stated for self-serve in the coverage I read.
Price trend: Digiday reported CPMs as low as $25 by mid-April, down from the $60 launch price.
Click-through: Similarweb’s panel puts the US all-time click-through rate at 0.73% (March 30 to September 16, 2026), with higher rates in newer markets such as Brazil (1.91%) on far less data.
Results so far: OpenAI published two partner-measured cases in October. For Dose, WorkMagic’s geo-lift test estimated 2.3 times more incremental orders than last-click attribution captured. For WeightWatchers, DV Rockerbox found attributed cost per acquisition 15.3% below the brand’s paid-search benchmark. Both were published by OpenAI, without sample sizes.
Independence: OpenAI says “advertising does not influence the answers ChatGPT provides,” and ads are “clearly labeled and separate.”

So an ad buys you a slot under the answer, never a mention in it. I’d test it only if your buyers already ask ChatGPT about your category, and measure with a holdout, as Dose did, not last-click.

What to do with this

Who this is for: businesses whose buyers ask an assistant to compare options. SaaS, online stores, agencies, B2B services, consumer brands people research. Who it’s not for: local services found mainly through Google Maps and reviews (fix your Business Profile first, as in the listing guide), and businesses whose buyers don’t research online.

Check you can be read (10 minutes). Open yourdomain.com/robots.txt and make sure OAI-SearchBot isn’t disallowed. The listing guide has the full check.
Measure your share (about an hour). Run the spreadsheet above: 10 questions, 5 runs each. Write down your share, the three brands named most often, and every source ChatGPT linked.
Map where the winners are talked about (about an hour). For each of those three brands, search Reddit and Google for “[brand] vs” and “[brand] alternatives”. List the subreddits, comparison pages and review sites that come up repeatedly, and the sources from step 2.
Fix your own facts (half a day). A public pricing page, a plain “what we do, for whom” page, and answers to the comparison questions from step 1.
Work the list (ongoing, a few hours a week). Answer questions in the two or three communities from step 3 under your own name, and say who you work for. Pitch the authors of the comparison pages. Ask happy customers for reviews on the review site that actually showed up for your category.
Re-measure monthly with the same questions. Expect slow movement: studies point at mentions that build up over months.
Skip hidden text, self-ranking listicles and “Summarize with AI” memory tricks.

What I’d do this week

Run step 2 and step 3. The list of places your competitors get talked about is your plan, and it costs nothing but two hours.

More on how assistants pick products and sources: How do AI shopping agents work? and glossary/share-of-model.

Sources

More of this every day on the channel: https://t.me/aidaily_business