How ChatGPT decides which brands to recommend

ChatGPT recommends brands using two sources of knowledge: what it absorbed during training, and, when it searches the web, what it finds and reads at the moment you ask. It does not consult a ranked list of businesses. It writes an answer, and the brands in it are the ones that its training and any retrieved pages made most prominent for that question. OpenAI does not publish a formula, so anything more specific than this is a guess.

Does ChatGPT use training or live search?

Both, depending on the question and the settings.

Training gives ChatGPT a broad, general memory of the world. It knows well-known brands, what they are known for and how people tend to describe them. That memory has a cut-off and cannot reflect recent changes.

Web search lets ChatGPT look things up. When it does, it issues one or more searches, reads the results and writes an answer informed by them, usually showing links to its sources. Training data or live search explains the split in more depth.

For a buying question such as "what is the best tool for X?", a search is common, because the answer depends on current options.

What shapes the brands it names?

Several things pull in the same direction.

Prominence. Brands that appear widely and consistently, described in similar terms across many sources, are easier for any language model to name. This favours established names, which is why newer brands often start with little visibility.

Fit to the question. Assistants try to match the need in the question. A brand described clearly as serving a specific use, such as agencies, small shops or a particular platform, is easier to match than one with a vague description.

What the search returns. If ChatGPT searches, the pages it reads matter a great deal. Roundups, comparison pages, reviews and forum discussions often carry more weight than a brand's own homepage, because they list several options and say who each suits.

How the question is phrased. "Best", "cheapest", "for beginners" and "alternatives to" all push the answer in different directions. The same brand may appear for one phrasing and vanish for another.

What does a typical recommendation look like?

When you ask ChatGPT for options, you usually see a short introduction, a list of a few named choices with a line on each, and often a note about how to decide between them. If it has searched, source links may sit alongside.

Three habits are worth knowing, because they affect how you count. First, it often hedges with "it depends on your needs", which means the same answer can name several brands without clearly favouring one. Second, it may list brands in an order that reflects nothing in particular. Third, it sometimes offers to narrow the answer if you add detail, such as budget or team size, which is a reminder that the first answer is only a starting point.

When you test, record the role each brand plays. A brand named as one option among five is in a different position from one the answer calls the best fit. Collapsing the two hides real differences.

Why does it sometimes recommend wrong or outdated brands?

Training memory goes stale, and web pages can be wrong or old. If the pages ChatGPT reads describe a product as it was two years ago, the answer may too. It can also confuse similar names, or merge a brand with its parent company. These are weaknesses of how language models handle entities, and they show up in real answers.

Is the answer the same for everyone?

No. Several things can differ between two people asking the same question:

• the exact wording;
• whether search is switched on;
• memory and custom instructions in a signed-in account;
• location, where it is passed along;
• ordinary variation between runs.

This is why one screenshot is weak evidence. Careful testing uses a clean session, repeats the question and counts. See how to test AI answers properly and how to check if ChatGPT recommends your business.

Does ChatGPT rank brands?

Not in the way a search engine does. A list in a ChatGPT answer may be ordered by relevance, alphabetically, by popularity or by no clear logic. Being named first does not necessarily mean being preferred, unless the answer says so in words such as "my top pick".

What can you actually do?

Nothing here is a trick, and results are not guaranteed. The sensible moves are the ordinary ones, done well:

1. State clearly who you serve and what you do, in plain words on your own pages.
2. Keep descriptions consistent across your site, profiles and listings, so that sources agree.
3. Earn a place on pages that list options, such as honest roundups and comparison content, since these are frequently read when ChatGPT searches.
4. Let the crawlers in. If your site blocks the bots that fetch pages for search, those pages cannot inform an answer. See AI crawlers and robots.txt.
5. Measure. Ask your customers' questions, without your brand name, on a regular schedule and note what comes back.

How do you know if it is working?

For a quick first look at where you stand, the Citewake checker is one place to start.

You compare like with like over time. Keep a fixed question list, ask it the same way each period, repeat each question, and count how often you are mentioned, recommended and cited. Citewake's methodology describes one disciplined way to do that, and a single answer should never be treated as the verdict.

What should you remember?

ChatGPT has no hidden leaderboard to climb. It reads, remembers and writes. The brands it names reflect how clearly and widely they are described, in memory and on the pages it retrieves. Test often, count patiently, and treat every single answer as one sample among many.