Should I track prompts that mention my competitors by name?

Competitor-named prompts are the closest thing you can test to a buyer building a shortlist. Someone who types "alternatives to [competitor]" already knows the category and is open to switching. Someone who types "[competitor] vs [you]" is choosing. If the answer leaves you out, you lose that comparison before it starts.

These prompts answer two questions the other groups don't:

• Do engines put you on the list when a buyer starts from a rival?
• Which pages do engines cite when they build that list?

Keep them apart from branded and unbranded prompts. A branded prompt names you, so you should appear almost every time. An unbranded prompt names nobody, so it tests category recall. A competitor prompt names someone else, so the engine starts from pages about them. Each group has a different baseline. Averaging them hides the one number you wanted.

Say you sell scheduling software and two rivals keep coming up in sales calls. A small, fixed set covers the useful cases:

1. "Alternatives to [Rival A]" and "Alternatives to [Rival B]".
2. "[Rival A] vs [Rival B]", with you absent from the prompt. This tests whether the engine adds you unprompted.
3. "[Rival A] vs [your brand]". This tests whether the engine describes you correctly in a head-to-head.
4. "Is [Rival A] a good fit for [your ideal customer]?"

Pick rivals your own size. A comparison against the biggest name in your category mostly measures their size. Brand prominence is a strong association in the evidence, and nobody controls it: a study of AI Search Arena data (arXiv 2507.05301, March to May 2025) found the 20 most frequent news sources took 67.3% of citations for OpenAI models. The method for benchmarking against like-sized competitors explains how to choose the set.

Keep the wording fixed between runs. Changing a prompt changes the measurement, and your trend line stops meaning anything.

The answer is half the value. The citations are the other half. When an engine builds a comparison, log every URL it cites, per engine and per run.

That list tells you where the comparison is being decided. The evidence shows a strong association: when engines cite other sites instead of yours, being named on those sites is the route into the answer. It is an association, not a promised effect. Working through prompts where competitors appear and you don't turns the citation log into a list of pages to approach.

Two cautions apply:

• Cited is not the same as used. OpenAI's API returns the sources it consulted separately from its citations, and the consulted set is often larger. Where an API exposes both, log both.
• Check the facts, not just the mention. Head-to-head answers are where engines state your price, features and plan limits. A wrong fact here costs you more than an absence.

If a third-party page misrepresents you, treat it as a real exposure. Some pages carry hidden text or injected instructions aimed at AI systems. Never answer with the same tactic: it is manipulation, and it puts your own pages at risk.

Competitor prompts are useful. They are also a sample, and the sample has known gaps.

• Your prompt is not the engine's query. OpenAI says ChatGPT search typically rewrites a question into one or more targeted queries. The Gemini API also generates its own search queries. Any prompt set is a proxy for an input you cannot see.
• Retrieval is discretionary. As of October 2026, ChatGPT decides per question whether to search, and web search in Claude.ai is a user toggle. An answer with no citations may come from training data with a fixed cutoff. Record whether each answer cited sources.
• Engines differ, sometimes sharply. Google says AI Overviews and AI Mode may use different models and techniques. In Otterly's 2,000-blog experiment, Microsoft Copilot produced 2,718 of the 3,934 citations tracked for one site while ChatGPT barely cited it. Score each engine on its own.
• There is no rank. No engine publishes position data for AI answers. Record named or not named, how you are described, and which sources were cited.
• Context changes answers. Memory and custom instructions can shape ChatGPT's search. Run logged out with memory off, as the reproducible ChatGPT tracking method describes. On Perplexity's Agent API, web search is opt-in, so set it explicitly.

1. Write six to ten competitor prompts against two or three like-sized rivals. Freeze the wording.
2. Run them on each engine you care about. Record whether you were named, how you were described, which URLs were cited, and whether the answer cited anything at all.
3. Rerun monthly and compare engine by engine, never as one blended score.
4. Fix what is in your control. Correct any wrong facts on your own pages. State specific facts (a price, a plan limit, a one-line definition) on a page crawlers can fetch, because that gives an engine a sentence it can quote. Check that your robots.txt lets OAI-SearchBot, PerplexityBot and Claude-SearchBot in, and that key facts sit in the HTML before JavaScript runs.
5. Contact the third-party pages engines cite most often and ask to be included where you genuinely fit.

None of this guarantees a mention. The engines decide what they say. You control whether they can reach accurate facts about you.

For a quick first look, the free way to see which competitors AI recommends is a starting point, and the free checker on this site shows whether engines name you for up to three buyer questions.

How many competitor prompts should I track? Start small. Six to ten prompts across two or three like-sized rivals is enough to see a pattern without drowning in data. Cover the main shapes: "alternatives to X", "X vs Y" without you, "X vs you", and a fit question for your ideal customer. Add more only when a new rival starts appearing in sales calls or in the cited sources.

Should my own brand appear in comparison prompts? Use both versions. A prompt without your brand, such as "Rival A vs Rival B", tests whether engines add you unprompted. A prompt with your brand, such as "Rival A vs you", tests whether they describe you correctly. The first measures inclusion. The second measures accuracy. They answer different questions, so keep them as separate rows.

What if one competitor's own site is the main source cited? That tells you the engine is building the comparison from the rival's version of events. Check that answer for wrong facts about you. Then make sure your own pages state the same facts clearly and can be fetched by AI crawlers. Also look for neutral third-party pages in the citation log, since being named on those is associated with appearing in the answer.

Do competitor prompts show my AI ranking against rivals? No. No engine publishes position data for AI answers, so any "AI rank" has nothing to be checked against. What you can record is whether you were named, how you were described and which sources were cited. Compare those counts engine by engine over time, and treat them as a sample of answers, not a league table.

How often should I rerun competitor prompts? Monthly suits most teams. Answers vary between runs, models change and retrieval fires at the engine's discretion, so one snapshot can mislead. A fixed monthly run with frozen wording gives you a comparable series. Rerun sooner after a rival launches a product or changes pricing, since that is when comparison answers are most likely to shift.