AI Search Visibility
Competitor AI Visibility Audit: Run One in an Afternoon
A step-by-step competitor AI visibility audit you can finish in one afternoon, with a scoring rubric, a gap decision table and a ranked fix list.

A competitor AI visibility audit compares which brands AI engines name, in what position and from which sources, for the questions your buyers ask. You can run a first pass in one afternoon: build 20 to 30 prompts, run each several times on a few engines, log mentions and cited URLs, then sort every lost prompt into a content, source or entity gap.
A competitor AI visibility audit is a structured comparison of which brands AI engines such as ChatGPT, Perplexity, Gemini and Claude name for your buyers' questions, in what position, and from which cited pages. You can finish a useful first pass in one afternoon with a prompt list, a spreadsheet and a few engines, and it ends with a ranked list of the gaps to close.
Key takeaways
- Audit the competitors the AI engines name, not only the ones you rank against on Google. Discover them with category prompts first.
- Run every prompt more than once. A single run is a snapshot of a volatile answer, not a measurement.
- Log five things per answer: brands named, position, cited URL, the wording used about each brand, and the engine.
- Sort each lost prompt into one of three gaps: content, source or entity. Each has a different fix, owner and timeline.
- Fix the gaps on high-intent prompts (comparison and category) first, then re-run the same prompts on a schedule to confirm the change.
- Manual audits show you the pattern; a tracker is what keeps it current after the afternoon is over.
What does a competitor AI visibility audit actually tell you?
The audit answers one question: for the prompts that matter to your buyers, why does a competitor get named and cited when you do not? It does not tell you how much revenue AI search will drive for your category, and it does not replace a wider marketing plan. It tells you what the engines are reading and repeating about your rivals that they are not reading or repeating about you.
That narrow scope is its strength. The output is a gap map you can hand to a writer, a PR lead or a developer on Monday morning.

Why is a competitor audit in AI search harder than in Google?
Three things make it harder. The competitors can be different, the answers change between runs, and the evidence sits in cited sources rather than in a ranking position.
First, the brands an AI engine names are often not the brands on Google's first page. Alhena AI, April 2026, reports Profound's finding that only 12% of URLs overlap between ChatGPT citations and Google's top organic results. If that holds for your category, your Google rival list is the wrong audit list.
Second, answers are unstable. The same Alhena AI article reports an AirOps finding that only 20% of brands stay present across five consecutive runs of the same query. A one-pass check can make a competitor look dominant, or absent, by chance. That is why the method below repeats every prompt.
Third, there is no rank number to compare. You compare mentions, order, wording and cited pages, which is why you need a log, not a glance.
How do the three gap types differ? A decision table
Every prompt you lose falls into one of three gaps. Naming them early keeps the audit from ending as a pile of screenshots. The table is the framework this guide uses from step 5 onward; it is a working model for prioritising, not a published benchmark.
| Gap type | What you see in the answer | Typical cause | Who owns the fix | Speed to fix |
|---|---|---|---|---|
| Content gap | The AI cites the competitor's own page, and you have no page that answers that question | Missing FAQ, comparison page, use-case page or specification detail | Content or product marketing | Fastest: you control it |
| Source gap | The AI cites third-party pages (reviews, listicles, forums, news) that name the competitor and not you | No presence in the sources the engine trusts for this topic | PR, partnerships, community | Slowest: you have to earn it |
| Entity gap | You are named, but described vaguely, wrongly or for the wrong audience; or the AI seems unsure what you are | Inconsistent brand description across your site and profiles | Brand and web owner | Fast to edit, slower for engines to reflect it |
The old rule of thumb that you should wait until one gap is closed before starting the next does not hold up. The three have different owners and can run in parallel, and the source gap is the one that benefits most from an early start because it has the longest lead time.
How do you run a competitor AI visibility audit in one afternoon?
The steps below take roughly three to four hours for one brand and two or three competitors. Each step states the action, the result and why it matters.
Step 1: Discover your real AI competitors (30 minutes)
Action: Write 10 to 15 unbranded category prompts, such as the questions a buyer asks before they know any brand name. Run them in fresh sessions with no history on at least three engines. Result: a list of every brand named. Why: the competitors the engines name may not be the ones you expect.
Alhena AI suggests treating brands that appear in half or more of the responses as your real AI competitors. Its arithmetic is simple: 10 prompts on 4 platforms gives 40 responses, so a brand needs to appear in 20 of them to cross that line. Take that as a reasonable working threshold, not a standard.
Step 2: Build the audit prompt set (30 minutes)
Action: Write 20 to 30 prompts across three intents: discovery (best options for a use case), comparison (you versus a named competitor, and competitor versus competitor) and purchase (a specific need with a price or feature constraint). Result: a prompt list tagged by intent. Why: intent tags let you weight the results later, because comparison and purchase prompts sit closest to a decision.
If you are unsure how many prompts are enough, our guide to how many prompts to track sets out the trade-offs, and which prompts to track explains how to pick them. Use the language buyers use, not your internal product names.
Step 3: Run each prompt several times on a few engines (90 minutes)
Action: Run each prompt three times on each engine you audit, in fresh sessions, logged out of any personalised account where you can. Result: three samples per prompt per engine. Why: because answers vary run to run, a single sample can mislead.
Personalisation can also shift what you see. Our note on personalisation in LLM responses covers how to keep your own history from skewing a test. Use the same wording every time, and write down the date and engine version if the product shows one.

Step 4: Log five data points for every answer (included in step 3)
Action: For each response, record the brands named, their order, every cited URL, the descriptive wording used about each brand, and the engine. Result: a flat table you can filter. Why: these five fields are what make the gap types visible; a yes-or-no mention column hides them.
| Column | What to enter | What it lets you answer |
|---|---|---|
| Prompt and intent | Exact text, plus discovery, comparison or purchase | Which stages of the buyer journey you lose |
| Engine and run number | For example Perplexity, run 2 | How stable the result is |
| Brands named, in order | List with position numbers | Who is recommended first, not just who is present |
| Cited URLs | Every link or source shown | Which pages feed the answer |
| Wording about you and rivals | Short phrase copied from the answer | Entity gaps: vague, wrong or missing descriptions |
Step 5: Calculate appearance rates and find the lost prompts (20 minutes)
Action: For each brand, divide the number of responses that name it by the total responses for that prompt group. Mark every prompt where at least one competitor appears more often than you. Result: a lost-prompt list. Why: it gives you a fair, repeatable score and focuses the next step on the prompts that matter.
The following worked example uses only the arithmetic from Step 1, with no data about any real brand. Say you test 10 prompts on 4 engines, three runs each, and log 120 responses. If Brand A is named in 54 of them, its appearance rate is 54 divided by 120, which is 45%. Brand B named in 78 is 65%. Brand B is the clear leader in the sample, and every prompt on which B appears and A does not goes on the lost-prompt list. The numbers are an illustration of the method, not a benchmark.
If you want to track this as share of voice over time, our AI search share of voice guide covers the calculation in more depth.
Step 6: Open every cited source on the lost prompts (45 minutes)
Action: For each lost prompt, open the pages the AI cited and classify each as the competitor's own page, a third-party editorial or review page, a community thread, or a reference page. Result: a source map showing where the competitor's advantage actually lives. Why: the engine's answer is built from those pages, so they show which fix will work.
Look for patterns across lost prompts rather than treating each one alone. If most citations are the competitor's own comparison and FAQ pages, you have a content gap. If most are third-party roundups that omit you, you have a source gap. For a deeper treatment of that pattern, see why a competitor is cited in AI answers and you are not.
Step 7: Classify, prioritise and assign owners (30 minutes)
Action: Tag every lost prompt as a content, source or entity gap using the decision table. Rank by intent first (comparison and purchase ahead of discovery), then by how often the competitor appears. Name one owner per gap type. Result: a ranked action list. Why: an audit without owners usually ends in a spreadsheet nobody opens again.

What should you check on your own site before blaming the competitor?
Some gaps are self-inflicted. Before you commit to fixes, check that the engines can reach and understand your pages. This short list takes about ten minutes.
- Confirm your robots rules do not block the AI crawlers you want to be read by. Our AI crawler user agents list shows what to allow and what to block.
- Check that the key answer sits in the visible text of the page, not only in images or scripts.
- Compare how your homepage, About page and product pages describe the brand. If they differ, fix the entity first; our brand description guide shows how to write one.
- Check external profiles and reference entries for the same wording. Wikidata and sameAs for brands explains how to link them.
- Run the GEO readiness scorecard for a quick read on the basics.
How do you turn audit findings into a fix list?
Match each gap type to a concrete action. The table is a starting point; adapt it to your category.
| Gap | First action | Evidence to re-check |
|---|---|---|
| Content: no page answers the question | Publish a page that answers the prompt in its first paragraph, with specifics a buyer would ask for | Whether your page begins to appear in the cited URLs for that prompt |
| Content: page exists but is not quoted | Rewrite the opening as a direct, self-contained answer with the brand named in full | Whether the answer wording starts to mirror your page |
| Source: absent from roundups and reviews | Pitch the specific publications and lists the engines cited, with facts they can verify | Whether the cited third-party pages mention you |
| Source: absent from community threads | Take part honestly where buyers ask, and follow each platform's rules | Whether threads you join appear in citations |
| Entity: vague or wrong description | Align the description across your site, profiles and reference entries | Whether the wording the AI uses about you changes |
For writing patterns that tend to be quoted, see the sentence structure that gets cited in AI answers and how to publish original data.
How do you know a fix worked?
Treat each fix as a small experiment. Change one thing, keep the prompt set unchanged, and re-run the same prompts on a schedule. Because answers vary, compare appearance rates across several runs, not one. Our GEO experiment design guide explains how to set up a before-and-after test and what counts as a result.
Keep the audit and the report separate. The audit tells you what to fix, while a monthly report tells leadership whether the share is moving. Our monthly AI search metrics report guide shows a layout that works.
What are the common mistakes in a competitor AI visibility audit?
- Auditing only your Google competitors. The engines may name different brands, so start with discovery prompts.
- Running each prompt once. Answers vary between runs, so one pass is a snapshot.
- Logging only yes or no. Position, wording and cited URL are what reveal the cause.
- Mixing branded and unbranded prompts. Branded prompts will always favour you; keep them in a separate group.
- Skipping the cited pages. The answer is the symptom; the sources are the cause.
- Treating the afternoon as the finish. A first pass has no trend line. Repeat it on a fixed cadence, or automate it.
When is a manual audit enough, and when do you need a tracker?
A manual audit suits a first look, a pitch to leadership, or a small category. It stops scaling when you add engines, markets, products or competitors, because every prompt needs several runs and every run needs logging. A tracker repeats the same prompts daily and keeps the history, so you see change rather than a snapshot.
If you are weighing tools, our comparison of AI search visibility tools and the pages on Noma versus Profound set out what to check before you buy.

How does Noma help with a competitor AI visibility audit?
Noma by Nurdd is an AI search visibility dashboard that tracks whether a brand appears in answers from 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, polled daily. It shows citation frequency by category, a competitor comparison and recommendations for improving visibility, so the afternoon's spreadsheet becomes a view that updates itself.
Book a demo at https://www.nurdd.club/demo to see your own brand and your named competitors in Noma: which engines name them, which pages they cite, and where your gaps sit.
Noma has a 3-day free trial. A card is taken at signup and billing starts on day 3; cancel before then and you are not charged. Noma is at nomahq.app. If you would rather have a team close the gaps, Nurdd's answer engine optimization agency and generative engine optimization services do that work, measured in Noma. For how the pieces fit, see how Nurdd works.
The afternoon checklist
- Write 10 to 15 unbranded category prompts.
- Run them on three or more engines in fresh sessions and list every brand named.
- Pick the competitors that appear most often.
- Write 20 to 30 prompts tagged discovery, comparison or purchase.
- Run each prompt three times per engine.
- Log brands in order, cited URLs and the wording used about each brand.
- Calculate appearance rates per brand.
- List the prompts a competitor wins and you do not.
- Open the cited pages and classify them.
- Tag each loss as a content, source or entity gap.
- Rank by intent, assign one owner per gap type and set a re-run date.
How we researched this
We reviewed the current top-ranking guide on competitive AI visibility audits, from Alhena AI, and rebuilt the method from it and from Nurdd's own guides on prompt design, share of voice and experiment design. The two figures quoted above are reported by Alhena AI as findings from Profound and AirOps; we did not open those original studies, so treat them as second-hand and check the originals before you cite them. Sources were checked on the publication date. We did not test any tool hands-on for this post, and the worked example is arithmetic only, not data about any real brand.
Frequently asked questions
How long does a competitor AI visibility audit take?
A first pass for one brand and two or three competitors takes about three to four hours. Most of that is running and logging the prompts. Repeating the audit is quicker once the prompt list exists.
Which AI engines should I include?
Start with the ones your buyers are most likely to use, usually ChatGPT, Perplexity, Gemini and Google's AI answers. Add Claude and others as you widen the scope. Different engines can name different brands for the same prompt, so more than one is needed.
How many prompts do I need?
Twenty to thirty prompts across discovery, comparison and purchase intent is enough for a first audit. Fewer than ten gives too little to see a pattern. More is useful once you are tracking over time.
Why run each prompt more than once?
Answers differ between runs. A competitor can appear in one run and vanish in the next, so repeated runs give you an appearance rate rather than a lucky or unlucky sample.
How often should I repeat the audit?
Repeat it on a fixed schedule, monthly for most teams, and after any major content or PR push. AI answers move, so a one-off audit goes stale. A tracker removes the manual work of repeating it.
Can I do this without paid tools?
Yes. A spreadsheet and free access to the engines are enough for the first pass. The limit is time and consistency, not tooling, which is where a tracker starts to earn its place.
What if the competitor has far more third-party coverage than I do?
That is a source gap, and it is the slowest to close. Start outreach early, target the specific pages the engines cited, and fix content and entity gaps in parallel while you wait.
Next step: see your competitors in Noma
An afternoon gets you a gap map. Keeping it current needs a daily view. Book a demo and we will show you your own brand and named competitors across the engines Noma tracks, with the pages each one is cited from. You can also run the free GEO readiness scorecard first, or browse all free tools.