Why ChatGPT recommends your competitor: 7 causes we check in every audit

A founder asks ChatGPT for the best tools in their category and gets a confident shortlist of three competitors. Their own product, which is cheaper and arguably better, is not there. It feels arbitrary. Most of the time it isn't.

When an assistant answers a "best tools for X" question with web search on, it retrieves a handful of pages, reads them, and summarises who those pages recommend. When it answers without search, it relies on what it learned in training. Either way, the answer reflects what is written about your category on the web, weighted towards sources the model trusts. The seven causes below are the ones we check in every audit, in the order we check them: the first two are technical and quick to fix, the rest take longer.

1. AI crawlers can't get in

Start here, because nothing else matters if this is broken. OpenAI says plainly that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. PerplexityBot plays the same role for Perplexity.

Two places to look:

• robots.txt. Search it for OAI-SearchBot, GPTBot, PerplexityBot, ClaudeBot and Bingbot, and for a blanket "Disallow: /" under "User-agent: *". Blocking GPTBot only stops OpenAI's training; blocking OAI-SearchBot removes you from ChatGPT search.
• Your CDN or firewall. Cloudflare has offered one-click AI crawler blocking since September 2024, and since July 2025 every new domain on Cloudflare chooses at setup whether AI crawlers get in. A site can have a perfect robots.txt and still return errors to AI crawlers at the edge. Check your bot settings, and check your logs for responses to those user agents.

2. Your content only exists after JavaScript runs

In a December 2024 analysis of crawler traffic, Vercel and MERJ found that none of the major AI crawlers render JavaScript: GPTBot, ClaudeBot and PerplexityBot fetch the HTML and stop there. Google's Gemini is the exception because it uses Googlebot's rendering.

If your pricing, feature lists or comparison tables are loaded client-side, those crawlers see an empty shell. The test is simple: open the page with JavaScript disabled, or run curl on the URL, and read what comes back. If the text you care about isn't in the raw HTML, assistants other than Gemini can't quote it.

3. No page answers the question the buyer asked

Assistants cite pages that answer the exact question. If buyers ask "best [category] for small teams" or "[competitor] alternatives", and your competitor has a page titled exactly that while you have a homepage and a features page, the competitor gets the citation.

Map the 10 to 20 questions your buyers actually ask, then check which of them have a page on your site that opens with a direct answer. Comparison pages, alternatives pages and use-case pages are the usual gaps.

4. You're missing from the pages assistants trust

This is the biggest cause and the slowest to fix. In a July 2026 study of 128 B2B software queries across Perplexity and OpenAI's search API, a vendor's own pages made up only 12 to 24 percent of citations. The rest came from independent sources: publications, listicles, review platforms and blogs. The single most-cited domain was TechRadar; G2 appeared across 13 of 16 categories.

So look at the sources, not just the answer. Run your buyer questions, collect every page the assistant cites, and check whether you are mentioned on each. If your competitor appears in eight of the ten articles an assistant reads and you appear in none, that is the answer to "why them". The fix is outreach to those authors, getting listed on the review sites that show up, and earning new third-party coverage, not more pages on your own site.

5. The assistant isn't sure what you are

Models group companies into categories. If your homepage says "the operating system for revenue teams", your LinkedIn says "sales engagement platform" and your G2 profile says "CRM", an assistant may not file you under the category the buyer asked about. Competitors with one plain, repeated description get matched more reliably.

Write one sentence that names your category the way buyers name it, and use it word for word on your homepage, About page, structured data, LinkedIn, review profiles and directory listings. Then ask each assistant "What is [your company]?" and see whether the answer matches.

6. You're absent from Reddit and YouTube

Assistants differ in where they look. In the same July 2026 study, YouTube and Reddit citations appeared only in Perplexity's answers, not in OpenAI's. If a competitor has a few useful Reddit threads and a YouTube walkthrough and you have neither, they will tend to show up in Perplexity and you won't.

This doesn't mean posting promotional comments. It means answering real questions in the communities your buyers use, from your own account, and publishing a short demo or comparison video.

7. You're newer than the model's memory

When an assistant answers without searching, it relies on training data with a cutoff. A company founded or renamed after that cutoff, or one that was barely written about before it, can be invisible in those answers regardless of what you fix today. Your older competitor has years of mentions baked in.

You can't change training data directly. What you can do is make sure the answers that do use web search can find you, and keep building third-party mentions so a future training run includes you.

How to run this check yourself

1. Write down 10 questions a buyer asks before they shortlist, in their words.
2. Ask each one in ChatGPT, Perplexity, Gemini and Google (for AI Overviews), and record who is named and which pages are cited.
3. Work through causes 1 and 2 on your own site. They usually take an afternoon.
4. For causes 3 to 6, compare your presence with the cited pages and the named competitors. That gap list is your plan.
5. Re-run the same questions in four to six weeks with the same wording.