How to Audit Your Competitors' Visibility in ChatGPT and Perplexity (Tactical Guide)

When a customer asks ChatGPT or Perplexity what options exist in your category, the answer names some brands and leaves others out. Knowing which ones appear, in what order and how they are described tells you a lot about your position against the competition.

This guide explains a method for manually auditing your competitors' visibility in ChatGPT and Perplexity, without tools, and when it makes sense to automate.

How it differs from classic SEO competitive analysis

In traditional SEO, competitive analysis has a known method: you check Google positions with Semrush, calculate keyword overlap and compare backlinks, with mature tools for each step.

AI answers are different. There is no list of results, but written answers. Small changes in the question produce different answers, and models change with their updates. Here you compare frequency, position and context within the answers.

Three practical consequences:

• It is not enough to look at your own brand. You need to measure the entire competitive universe in each prompt.
• The key metric is not individual position, it is relative share of voice: what percentage of the relevant answers in my category include me versus how many include competitor X?
• Sentiment matters as well as presence. Appearing often with negative descriptions is a different problem from not appearing.

Three things you need to know before you start

1. Your competitor set is not your commercial hit list. Your competitors in GEO are the brands that appear alongside yours in AI answers. Sometimes they are not the ones your sales team considers competition. A new brand that was not on your list may appear. You only discover that by measuring.

2. Same prompt, different session, different answer. Answers vary between runs and, if you use an account with memory or history enabled, those can influence them. For auditing, use sessions without history (temporary chat, a clean account or a signed-out session) and note the conditions.

3. One model does not give you the full picture. ChatGPT, Perplexity, Gemini and Claude answer differently. If you only audit ChatGPT, you have a partial snapshot. Cover at least three models — learn the differences between ChatGPT, Gemini and Perplexity before you start.

10 prompts to manually audit your competitors

Replace [your category] with your vertical (CRM, SEO agencies, accounting software, boutique hotels in Austin — whatever applies) and [your brand] with your company name. Run each one in a session without history, in both ChatGPT and Perplexity.

Prompt 1. Pure discovery.

"What are the best [your category] in 2026?"

What you're looking for: a neutral list. Note which brands appear, in what order, and whether yours is included.

Prompt 2. Three-way comparison.

"Compare [your brand] with [competitor 1] and [competitor 2]. Which is better for [typical use case]?"

What you're looking for: how the AI describes you against those two rivals. Sentiment, strengths, weaknesses.

Prompt 3. Use-case recommendation.

"I'm [typical buyer profile] and I need [specific problem solved]. Which [your category] do you recommend?"

What you're looking for: your position on the list for your exact ICP. This is where you discover whether the AI recommends you to your ideal customer.

Prompt 4. Pricing.

"Which [your category] is the most affordable, and which is the premium option?"

What you're looking for: does the AI classify you correctly within your price range, or does it place you somewhere you don't want to be?

Prompt 5. Alternatives to the leader.

"Give me 5 alternatives to [category leader]"

What you're looking for: if you appear in the list of alternatives to the leader, you are considered a real option. If not, you do not appear in that answer for people looking for alternatives to the leader.

Prompt 6. Reverse angle.

"Give me 5 alternatives to [your brand]"

What you're looking for: the AI tells you who your competitors are in its model of the world. Sometimes brands you have never tracked appear here — a shortcut for competitor discovery that wasn't on your map.

Prompt 7. Geography.

"Which [your category] is best for companies in the US / UK / [your market]?"

What you're looking for: if you sell locally, this prompt reveals whether the AI associates you with the right market or ignores you in favor of global brands.

Prompt 8. Feature-specific.

"Which [your category] has the best [your key differentiating feature]?"

What you're looking for: whether the AI associates you with the feature that is your differentiator. If not, check whether that differentiator is clearly explained on your website and in external sources.

Prompt 9. Reviews.

"What do users say about [your brand]? And about [competitor 1]?"

What you're looking for: sentiment. Compare positive versus negative framing in both cases. Identify which source the model is drawing those opinions from.

Prompt 10. Active citation.

"Recommend a [your category] and give me a link to their website."

What you're looking for: citation rate. If the AI recommends your competitor with a link and you without one, review which pages it cites, whether your website has an equivalent page, your schema markup and your presence in external sources.

A basic manual audit is these 10 prompts × 3 models × 2 repetitions = 60 runs. Record the date, model, language and country of each one.

Separate two groups when analyzing. Prompts 2, 6 and 9 name your brand: they show how the AI describes you, but they do not measure whether you appear spontaneously. Prompts 1, 3, 4, 5, 7, 8 and 10 do not name it: they are the ones that measure discovery, with 7 × 3 × 2 = 42 runs as the denominator.

How to interpret the results: a 4-quadrant framework

With the results of the discovery prompts, classify each competitor on two dimensions:

X-axis: appearance frequency (in how many of the 42 discovery runs they appeared)

Y-axis: quality of appearance, best read as two separate data points: sentiment (from −1 to +1, higher is better) and average position (lower is better). Do not add them together: they are on different scales and run in opposite directions. If you want a single axis, define a normalized formula beforehand and explain it; otherwise, chart each one or use whichever matters most for your decision.

Four quadrants:

• Established leaders (high frequency + good quality): study which sources describe them and how.
• Volume with weak descriptions (high frequency + low quality): they appear often but with criticism or at the end of lists. Review what the answers repeat and whether your brand can offer a better-described alternative.
• Occasional but strong (low frequency + good quality): they appear rarely but well. They are often specialized brands; study what they do.
• Barely visible (low frequency + low quality): they hardly appear in these answers. That does not mean they are not relevant competitors in your market.

Where do you sit? That is the first question. The second: what change could move you to another quadrant, and how you will check it in the next measurement.

From manual to automated: when to make the jump

Manual auditing works for an initial snapshot and for validating hypotheses. It can also be repeated with a documented protocol, although it takes time. Three signs that it makes sense to automate:

1. You want a trend, not a snapshot. To know whether you are moving up or down you need to run the same questions at regular intervals and compare. With many questions and models, doing that by hand takes many hours.

2. Your category changes often. If new competitors enter frequently or you notice changes in answers, the snapshot ages quickly and you should measure more often.

3. You want to detect position changes. If a competitor goes from position 1 to position 5, something has changed (content, reviews, sources or the model itself). Detecting that requires comparing repeated measurements.

When you automate, expand the questions, cover the models relevant to your market and keep a fixed frequency. See Mentio's competitive radar.

5 mistakes that invalidate a competitive analysis in AI

1. Auditing with history. If you run the prompts from your personal ChatGPT account with memory or history about your brand, results can be biased. Use sessions without history.

2. One prompt per idea. "Best CRM?" and "What's the best CRM for my business?" return different answers. Audit variants, not just the canonical phrasing.

3. Ignoring the second paragraph. Long answers often include nuances and alternatives further down, and brands that are not in the first list can appear there. Read the whole answer.

4. Mixing languages. ChatGPT can cite different brands in English and in Spanish. If your market is English-speaking, audit in English.

5. Auditing without a baseline. Without a first measurement and a fixed prompt set, you cannot know whether you improved. Define your baseline before making any changes to your content.

Hypothetical example: an SEO agency that does not appear for its ideal customer

This example is illustrative; it does not describe a real case. A B2B SEO agency based in Madrid runs the 10 prompts in ChatGPT, Perplexity and Gemini.

Audit result: across the 42 discovery runs it appears in 5, in positions 5–7. Its three main competitors appear in 25–31, several in positions 1–2. The answers describe them as "technical SEO specialists for mid-sized companies in Spain"; the agency is not associated with that specialization, even though it is its core service.

A hypothesis worth checking: when reviewing the cited sources, the competitors have many pages on technical SEO and mentions in industry media; the agency has a few scattered articles. That is a correlation that guides the work, not proof that the amount of content causes visibility.

Possible actions: pages that clearly explain its specialization, Organization schema, documented cases and relationships with industry publications. Then repeat the same audit at regular intervals to check whether anything changes, without attributing any improvement to a single action.

A rival gaining ground does not by itself explain your decline: when you diagnose a ChatGPT visibility drop, require cell-by-cell coincidence and an observable mechanism before attributing the drop to that competitor.

When one-to-one auditing is not enough, the next step is moving from a single rival to a peer population with the AI visibility industry benchmark.

FAQ

Is it legal to audit competitors in ChatGPT?

Asking a public assistant questions and reading how it describes other brands is what any user would do; you are not accessing private data. If you automate queries, review each service's terms of use and use their APIs where appropriate.

How many times should I repeat each prompt?

At least twice in separate sessions, and more if answers vary a lot. The same prompt can produce different answers; repeating helps you tell a stable pattern from noise.

Is running the audit only in ChatGPT enough?

It works as a starting point, but it is incomplete. Each assistant uses different sources and your customers may use several. Check with your data (referrers, surveys) which ones matter in your sector.

How quickly can a competitor's AI visibility change?

There is no fixed timeline: it depends on the model, whether it uses web search and what changes in the sources. If you see a sudden change, investigate what the competitor published or what coverage it earned, and repeat the measurement before drawing conclusions.

What should I do if a competitor far outpaces me in share of voice?

First, check whether it really competes with you or has simply been on the web longer. Then review which sources the AI cites when it talks about them: their own content, media, reviews, directories. That gives you hypotheses about what to work on in your category.

Can I automate this with custom scripts, or do I need a dedicated tool?

You can do it with the APIs of OpenAI, Perplexity and other providers. You will need to build the execution, logging, brand detection and maintenance. Compare that development cost with the price of a tool, depending on how many brands you measure.

Starting your competitive audit with Mentio

Mentio measures the visibility, position and sentiment of your brand and your competitors in ChatGPT, Claude, Gemini, Grok and Perplexity, depending on the models in your plan, and shows how they evolve between analyses you run on demand. It does not send automatic alerts.

See the competitive radar →

Want to know if AI mentions your brand?

Discover your visibility in ChatGPT, Claude and Gemini in minutes.

Analyze your brand for free

Related articles