How ChatGPT Picks Which Brands to Recommend (and How to Get on the Shortlist)

• To get recommended by ChatGPT, your brand needs to be on the shortlist the model holds before it searches. In one network-capture study, brands ChatGPT wrote into its first search query were named in 68.9% of final answers, against 2.1% for brands it only found in the results.
• The shortlist is fed by entity recognition and repeated independent coverage: articles, reviews, directories and reference pages that name your brand in the context of your category.
• Your own website is the second step: once you are on the list, clear, crawlable pages decide whether ChatGPT can confirm, cite and describe you correctly.

You ask ChatGPT for the best tools in your category and it names three competitors, while your site ranks on page one of Google for the same phrase. The gap becomes predictable once you see the order in which ChatGPT works.

As of October 2026, the clearest public evidence comes from a network-capture study published in September and a sharp shift in ChatGPT’s citation sources in mid-August. Both point to the same conclusion: most of the recommendation is settled before the model reads a web page.

Does ChatGPT decide which brands to recommend before it searches?

The best available evidence says it does: ChatGPT appears to hold a shortlist of brands before it runs a web search, and brands on that shortlist dominate the final answer. On 14 September 2026, Tom Dehnel summarized research by Suganthan Mohanadasan, who captured the network traffic behind ChatGPT answers and split the brands into two groups: those ChatGPT wrote into its own first search query, and those it only met in the retrieved results. The first group was named in 68.9% of final answers and the second in 2.1%.

The same captures show that on 8 August 2026, site: queries jumped from 0.37% of ChatGPT’s query mix to 16.8% in a single day. A site: query restricts the search to a domain the model has already chosen, which only happens when the brand name is already in mind.

Treat this as one study. The summary discloses no sample size, and nobody outside OpenAI can trace the model’s preferences to specific sources. Its direction matches the citation data below.

What gets a brand onto ChatGPT’s shortlist?

Two things feed the shortlist: whether the model recognizes your brand as a distinct entity in your category, and how often independent sources name you in that context. When many unrelated pages describe a company as a payroll platform for startups, the association becomes strong; when only the company’s own site says so, it stays weak. Dehnel’s practical advice follows: if most of your AI search budget goes to tweaking your own pages, move part of it to earning coverage on independent sites.

The August change points the same way. OtterlyAI tracked 16 brand reports and found that ChatGPT cut Reddit citations by 73.4% in mid-August 2026, from 497 to 132 per day, while its total citation volume rose 3.5%. We covered the details in Reddit after ChatGPT’s August change.

Does your own website still matter for ChatGPT recommendations?

Your website matters as the second step: it rarely puts you on the shortlist by itself, but once you are there, your pages decide whether ChatGPT can confirm, cite and describe you correctly. The jump in site: queries shows the model now visits named domains to check them. It needs to find in plain text what you do, for whom and where. Information locked in images or late-loading scripts leaves it with a vague description or a competitor’s clearer page.

Add-ons such as llms.txt help only after the basics are in place: server-rendered text, an About page that states facts, product pages that answer buyer questions, and consistent Organization schema. The same pages feed Google’s AI answers, so it is worth running a separate test to check whether you appear in AI Mode.

What is a realistic 90-day plan to get recommended by ChatGPT?

Spend the first month fixing how your brand is described, the second earning independent mentions, and the third testing and tightening your own pages.

In month one, write one description sentence with your name, category, audience and market, for example “Acme is a payroll platform for UK startups with 10 to 200 employees.” Use it on your site, LinkedIn, Crunchbase, review profiles and founder bios, add Organization schema with sameAs links to those profiles, and fix outdated names. Consistency lets a model treat those mentions as one entity.

In month two, give independent sites reasons to mention you. The strongest is an original data asset, such as a benchmark or index with a clear method, pitched to trade media, newsletters and podcasts. Add customer case studies on customers’ own sites and brief the authors of independent “best X for Y” lists.

In month three, run the prompt test below, note which sources ChatGPT cites when it names competitors, and target those first. Background knowledge may change only when the model is updated, so judge the first 90 days by direction: more accurate descriptions, more cited sources that mention you, and a few new appearances.

How do you test whether ChatGPT recommends your brand?

Run the same 20 prompts every month in a clean session and record whether your brand appears, its position, which competitors are named and which sources are cited. Answers vary between runs, so run each prompt twice and count a mention only when it appears in both.

Log the date, the model shown and every cited source. After three months you will know which sources ChatGPT trusts in your category.

How do you get recommended by ChatGPT when your brand name is a generic phrase?

A generic name makes entity recognition harder because the model has to separate your company from the everyday meaning of the words, so every signal must repeat the same name, description and domain. Our own brand is an example: “Content Managers” is also a job title.

On content-managers.com we treat this as an entity problem. The site carries consistent Organization schema with sameAs links to our official profiles, so the domain, company name and profiles read as one entity. Articles carry a Person entity for Dror Feldman, the founder, connected to that organization. We also build data assets only we publish, such as the Israel Tech Hiring Index, which counts open roles on Israeli tech companies’ own job boards. Anyone who cites it has to name Content Managers in a context no job title shares. We report the steps here, with no claim about results; the prompt test above is how to judge them.

Frequently asked questions

How long does it take to get recommended by ChatGPT?

Plan for several months. Fixing descriptions and schema takes a few weeks, and earned coverage builds over a quarter or more. Part of ChatGPT’s knowledge changes only when the model itself is updated, on a schedule you do not control. Answers that draw on live search can shift sooner, so track both with a fixed monthly prompt set.

Does ranking on Google help you get recommended by ChatGPT?

It helps indirectly. Strong rankings make your pages easy to find when ChatGPT verifies a brand, and the links behind them also build recognition. Rankings alone rarely put you on the shortlist: in the study above, brands found only in retrieved results were named in 2.1% of answers.

Is Reddit still worth it for ChatGPT visibility?

Reddit became a weaker lever for ChatGPT in August 2026, when OtterlyAI measured a 73.4% drop in ChatGPT’s Reddit citations. In the same data, Reddit citations in Google, AI Mode, Gemini and Perplexity stayed roughly flat, so Reddit keeps a role on those engines and for reputation. For ChatGPT, official pages and independent coverage now carry more weight.

Why does ChatGPT describe my company incorrectly?

Usually because the sources it relies on disagree or are out of date. Old directory listings, a previous company name or a founder bio with outdated positioning can all feed the answer. Find the source of the error in the cited links of your prompt log, correct those pages first, then make your About page state current facts plainly.

Where should you start this week?

Write your one-sentence brand description and run the 20-prompt test once, so you have a baseline before you change anything. The baseline shows who owns the shortlist today and which sources put them there. If you want a second pair of eyes on it, our search and AI visibility team does this work for B2B companies.

Sources

• Tom Dehnel, ChatGPT Decides Which Brands to Recommend Before It Runs a Search
• OtterlyAI, ChatGPT cut Reddit citations by at least 73% in August 2026. Here is what replaced it.