What to Do When AI Recommends Your Competitor Instead of You

Why does AI recommend your competitor?

Usually because the sources the assistant trusts describe your competitor more clearly, more often or more recently than they describe you. AI assistants don't pick favourites at random; they summarize what they find.

When an assistant answers "What is the best tool for X?", it is drawing on two things: what the underlying model learned during training, and, for many questions, pages it retrieves from the web at the time of the question. If your competitor appears in more of those pages, in clearer terms and in the right context, it has a better chance of being named.

Common reasons a competitor wins the answer:

• They are on the comparison lists. "Best of" articles, roundups and directories that name them but not you.
• They have more third-party mentions. Reviews, forum threads, press and partner pages that repeat their name next to the category.
• Their own pages answer the question directly. A clear page that says what they do, who it's for and how it compares is easy to quote.
• Their positioning is consistent. When every source describes them the same way, the assistant can summarize them confidently.
• Your information is missing or outdated. If key facts about you are hard to find, the assistant leaves you out or describes you vaguely.

None of these are mysteries, and most of them can be checked. The mistake is to treat a lost answer as a branding problem you can only fix with more ads. It is usually a source problem, and source problems have concrete fixes. Start by looking at competitor benchmarking data across the questions that matter, so you know where you lose and to whom.

Which sources is the AI relying on?

Look at the citations behind the answers where your competitor wins. The domains that appear again and again are the ones shaping your category, and they are where your work should start.

Assistants that search the web often show their sources. Collect them across many answers rather than one. Patterns appear quickly: a few review platforms, a couple of industry publications, a handful of comparison articles, some community threads, and the competitors' own sites.

Group the sources into types, because each type calls for a different action:

• Your own site. Are any of your pages cited? Which ones? If none are, your site may lack clear, quotable answers to the questions buyers ask.
• Competitor sites. If competitors' pages are cited for general category questions, they have published the explainer you haven't.
• Editorial and comparison content. Lists and reviews written by third parties. Being missing here is one of the most common reasons for being missing in the answer.
• Community and user-generated content. Forums and Q&A sites where real users discuss options. These often carry strong weight for "what do people actually use?" questions.
• Directories and profiles. Listings that define what a company is and does.

Citation tracking in Refora shows which domains and URLs are cited across your tracked questions, and where competitors are cited and you are not. That turns a vague "AI likes them better" into a specific list of pages to work on.

What should you change first?

Start where the gap is largest and the fix is most within your control: missing answers on your own site, outdated facts about your brand, and the two or three third-party sources that appear most often in answers you lose.

A practical order of work:

1. Fix your own facts. Make sure your site states clearly what you do, who it's for, where you operate and how you differ. Update outdated descriptions on your profiles and listings. This is fast and fully in your control.

2. Publish the missing answers. If buyers ask a question and your competitor's page is cited, write a better, more direct page that answers it. Lead with the answer, then the detail. Make it easy to quote.

3. Earn a place on the sources that matter. Identify the comparison articles and review sites that appear most often. Reach out, provide accurate information, and get listed where it is legitimate to be listed.

4. Show up in real conversations. Where community threads shape answers, participate honestly. Answer questions, share expertise and avoid anything that looks like spam.

The Citation Agent in Refora turns these findings into concrete actions, with a brief for each one, so the work is specific rather than a general instruction to "do more content". Prioritize a few actions, finish them, and measure before moving on.

How do you know if the change worked?

Re-check the same questions in the same assistants after the change, and compare your mention rate, position and description with the period before. If you don't re-check, you can't separate real progress from normal variation.

This is the step most teams skip, and it is the one that makes the whole effort credible. The loop looks like this: data, action, re-check.

• Data. Track a fixed set of questions across ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews and Mistral, and record who is named and which sources are cited.
• Action. Make a specific change based on that data, and note the date.
• Re-check. Ask the same questions again over the following weeks. Compare visibility, share of voice and sentiment against the baseline.

Be patient and honest about timing. Assistants that search the web can pick up a new page quickly; changes in what a model learned in training take much longer. Look at trends over several runs, not one answer.

When a change works, you'll usually see it in more than one place: a higher share of answers that mention you, your page appearing among the citations, and a description that matches your positioning. When it doesn't, the data tells you that too, and you move to the next action.

That is how you go from "AI recommends our competitor" to a clear plan, and from a plan to proof.

Frequently asked questions