Why ChatGPT recommends your client's competitor — and how to explain it

"Why does ChatGPT recommend them and not us?" is the question that brings most clients to the AI visibility conversation. The honest answer is never "the algorithm". It's almost always one of a handful of concrete causes — and you can usually find out which by reading the answers and the sources behind them.

First: is it a pattern or one answer?

Before diagnosing anything, check the size of the problem. A client who saw a competitor in one ChatGPT answer has seen one draw from a distribution; the next answer might name them. Ask the question repeatedly — ideally the full prompt set, over weeks — and look at the rates: "the competitor is mentioned in 40% of answers, the client in 30%" is a pattern worth explaining. Why ChatGPT gives different answers every time covers the background.

The seven usual causes

1. The competitor is in the sources the engine reads

When ChatGPT, Gemini or Perplexity search the web, they lean on pages that already compile options: "best X" articles, review sites, directories, comparison pages, forum threads. If the competitor appears in those pages and the client doesn't, the engine has nothing to recommend the client from.

How to check: collect the cited domains in answers where the competitor is mentioned and the client isn't. Visit the top five. Is the client listed?

2. The competitor's site answers the exact question

AI answers favour pages that match the question directly: a "pricing" page that states prices, a "vs" page comparing two options, a city page for a local service, a use-case page for a specific audience. If the competitor has the page and the client doesn't, the competitor gets cited.

How to check: for each gap prompt, search the competitor's site and the client's site for a page that answers it.

3. The engine isn't sure what the client is

If the client is described differently across its website, profiles and press — a "studio" here, an "agency" there, a "platform" elsewhere — engines may not connect it to the category in the question. Competitors with a consistent, simple description get matched more easily.

How to check: ask the engines directly, "What is [client]? Who are its competitors?" If the answer is vague or wrong, entity clarity is part of the problem.

4. History and coverage

For answers given without a live search, the model relies on what it learned in training. Brands with years of coverage — major publications, Wikipedia, widely linked reviews — have an advantage that new brands don't, and it changes slowly.

5. Search visibility in the underlying engine

Gemini can ground answers in Google Search, so a competitor that outranks the client in Google for the question often also wins in Gemini. Perplexity uses its own search, ChatGPT its own. Weak classic SEO for the question tends to show up somewhere in AI answers too.

6. The engine can't read the client's site

If the client's robots.txt blocks the crawlers AI search relies on — for ChatGPT search, that's OAI-SearchBot — its pages can't be used in those answers. Also check that key content isn't hidden behind scripts, logins or interstitials.

7. The question is framed around a different category

Sometimes the client simply isn't what the question asks for: a premium service in "cheapest" prompts, a national brand in local prompts. That's not a visibility problem, and the right response is to adjust the prompt set, not the client.

A 30-minute diagnosis

1. Pick the three prompts with the biggest gap (client absent, competitor present).
2. Open five answers for each across engines. Note who is named and what reasons the engine gives.
3. List the cited domains and check whether the client appears on them.
4. Check the client's site for a page that directly answers each prompt.
5. Ask each engine "What is [client]?" and compare with how the client describes itself.
6. Check robots.txt for AI crawler blocks.

You'll usually end with one or two dominant causes — and a short list of fixes.

What to say to the client

Keep it concrete and evidence-led:

Avoid promising a position. You can't control what an engine says in a given answer; you can change the sources and pages it reads, and measure whether the rate moves. That's an honest promise, and the one clients renew for. For the delivery side, see GEO deliverables and how to prove ROI.

Make the evidence easy to show

The meeting goes better when you can open the actual answers. Ranxii stores every answer for every client prompt, highlights the client and competitor mentions, lists the cited sources, and shows the prompts where the client is absent but competitors appear — so steps 1–3 of the diagnosis are already done when you open the client.