Why ChatGPT Recommends Your Competitor (and How to Find the Sources Behind It)

When ChatGPT recommends your competitor instead of you, the reason is nearly always retrievable: the sources the engine reads at the moment of the question mention them and say little or nothing about you. The model is summarising the web's evidence about your category, and right now the evidence points next door. Closing that gap is the discipline called generative engine optimization (GEO), which you will also hear called answer engine optimization (AEO), and the useful first step is diagnostic: find the exact sources behind the recommendation. This piece shows you how to do that by hand, in roughly an hour, with ChatGPT itself and your own robots.txt file.

I have mapped the general mechanism, training data versus live retrieval, in how ChatGPT chooses which brands to recommend, so the theory here is one sentence: when a question benefits from current information, ChatGPT searches the web, reads what it fetches, and composes the answer from what those sources say. The competitor question is narrower. Why do the retrieved sources keep producing the other name? Across the audits I run, five findings cover it.

The five reasons the other name keeps coming up

1. They are in the lists the engine retrieves. Ask an assistant for the best anything and a large share of what it fetches is other people's shortlists: best-of roundups, comparison posts, directories. The engine wants to act as an objective broker, so it leans on sources that have already done the comparison work. If your competitor sits in six of those lists and you sit in none, the recommendation was settled before the question was asked.

2. They have more third-party corroboration. Every business describes itself as the best, which makes self-description nearly worthless as a signal, so the engines weigh what independent sources say about who you serve and what you do. This layer responds to deliberate work: building it for BizScout, a business marketplace, produced 450+ AI mentions over a year, and ChatGPT now cites them as the go-to source for business buyers.

3. Their site is readable by the engines and yours may be blocking them. OpenAI's crawler documentation is plain on this: sites that disallow OAI-SearchBot in robots.txt will not be shown in ChatGPT search answers, though they can still appear as navigational links, and GPTBot is a separate setting covering model training. Clients regularly arrive on GEO plans with AI crawlers blocked by default, usually through an old bot-management rule nobody has reviewed. The cheapest fix in this discipline.

4. Their pages are easier to lift. Retrieval systems quote passages that answer cleanly: a self-contained definition, a direct claim with a number attached, a comparison that states who each option fits. A page that buries its answer under eight hundred words of brand story gives the engine nothing safe to extract, and the citation goes to the competitor page that answered in its first paragraph.

5. The category consensus formed around them. When many independent sources agree on who the main options are, the engine inherits the agreement. This is rough on quiet market leaders who win through referrals, because the engine cannot read revenue. Visibility is the gap I find in most of these audits.

Grace Frank covers the same diagnosis from the business owner's side in her video on why ChatGPT recommends your competitor:

<YouTube id='0XjkoBVoI6M' />

The workflow: finding the sources behind the recommendation

You can buy tooling for this, but the first diagnostic costs nothing beyond an hour and a spreadsheet.

Step one: ask with search on. Put the question your buyer asks to ChatGPT with its search function enabled, phrased the way a buyer would phrase it. When the answer uses search it includes inline citations you can open, and a Sources panel beneath the response listing the pages it drew from, per OpenAI's documentation of ChatGPT search. Read every citation and note which pages put your competitor's name in front of the engine.

Step two: repeat across phrasings. OpenAI documents that ChatGPT typically rewrites your question into one or more targeted queries, sends them to search partners such as Bing, and may follow with narrower queries after reviewing the first results. Your buyer's words fan out into searches you never see, so one run tells you little. Ask the category question five or six ways and log every cited domain each time: best option for the use case, alternatives to the competitor, the comparison question, the local variant.

Step three: tally the recurrences. After ten or fifteen runs a pattern emerges, and in my experience it is a short one: a handful of domains supply most of the citations for any category. Those recurring domains are the battleground, because the AI answer is downstream of them. Your problem has stopped being "ChatGPT prefers my competitor" and become "these nine specific pages mention my competitor and ignore me", and a problem with page-level coordinates is a problem you can work.

Step four: check your own crawl access. Load yoursite.com/robots.txt and look for disallow rules naming OAI-SearchBot or GPTBot, and for blanket rules that catch all bots. Check your CDN and firewall settings too, since bot protection often blocks AI crawlers without a robots.txt entry. OpenAI notes it can take roughly 24 hours after a robots.txt update for its systems to adjust, so a fix registers quickly. The guide to AI crawler access and llms.txt covers anything suspicious you find here.

Step five: map the gap. For each recurring source from step three, record two columns: competitor present, you present. The finished sheet is your work queue, ranked by how often each source gets cited, and it usually contains a few industry roundups, a directory or two, a review platform, and sometimes a Reddit thread that outranks everything else. This mapping stage is the spine of our own visibility audit.

What to do with the map

The map converts directly into a programme. Pitch the roundups you are absent from, claim the directory listings, earn the reviews, and publish comparison content on your own site that does the broker's work for it, including stating plainly who you are the wrong fit for. Candour reduces the model's uncertainty about where you belong. Alongside the outreach, restructure your key pages so each buyer question gets a direct, extractable answer; the mechanics are in how AI citations get built.

Set expectations in months. New coverage has to be published, crawled, and retrieved before it shifts a single answer, and the fastest result we have published took five: Tides Mental Health, a behavioural health provider, moved from page 2 to page 1 with 4.7x organic traffic growth in that window, and AI engines now recommend them for anxiety treatment. The wait prices in well against the numbers we publish and plan around: organic search traffic is down roughly 30% across the board, and AI-referred visitors convert at 3 to 6x the rate of traditional organic. A buyer arriving from an AI recommendation arrives largely pre-sold, which is why the competitor holding that position hurts.

What this diagnostic cannot tell you

The same question on different days can cite different sources and name different brands; retrieval shifts and models sample, so treat any single answer, flattering or painful, as one roll rather than a verdict. Trends across many prompts are the unit of measurement. Personalisation adds a layer you cannot see: OpenAI documents that when Memory is enabled, ChatGPT may use what it knows about the user to sharpen the search query, so two buyers asking the same thing can trigger different retrievals. Model updates reshuffle the training-side baseline on a schedule nobody outside OpenAI knows. And OpenAI states directly that there is no way to guarantee top placement in ChatGPT search, so anyone selling a guaranteed recommendation is selling something the mechanism does not support.

Frequently asked questions

How do I get ChatGPT to recommend my company?

Work the three layers the diagnostic surfaces: make sure OAI-SearchBot and GPTBot can crawl your site, structure your pages so each buyer question gets a direct answer the engine can lift, and build third-party coverage in the sources the engine already cites for your category. The third layer moves the most and takes the longest, since every citation has to be published and crawled before it can be retrieved. Plan in months and measure trends across many prompts.

Should I just bid on my competitor's keywords instead?

Paid has a genuine place, especially when you need leads before earned visibility can mature, and competitor terms can be part of that. The trade is structural: you are paying to play consistently, and compounding the lead funnel means compounding the spend, while earned source coverage keeps answering buyer questions at one month, six months, and next year. I treat paid as the bridge and the source map as the asset; GEO versus SEO covers how the organic side splits.

Why does ChatGPT name a different competitor each time I ask?

Because answers are assembled fresh from retrieval each run, and retrieval results shift between runs. The engine also rewrites your question into search queries that vary with phrasing and, when Memory is on, with what it knows about the asker. What matters is the share of runs that include you. If three competitors rotate through the answers and you never appear, the rotation is noise and the absence is the signal.

Can I ask ChatGPT directly why it recommended my competitor?

You can ask, and it will produce a fluent explanation, but treat that explanation as a guess rather than a log file, because the model has no reliable access to its own reasons. The citations and Sources panel from a search-enabled answer show which pages the answer was composed from, so interrogate the sources rather than the model's account of itself.

The full diagnostic runs in an hour, and the result is a named list of pages deciding your category's AI answers. For the one-minute version of step one, run the free AI Visibility Check: it tells you whether ChatGPT names you at all, and either answer makes the rest of the workflow sharper. Start there, work the map, and let me know what the sources say about you.