An AI visibility audit should explain why a competitor appears in an answer and what evidence supports that appearance. A screenshot of one favorable response cannot do that. Results vary by engine, prompt, location, account state, and date. The audit needs repeated observations, source tracing, page comparison, entity analysis, and a clear separation between mention, citation, and business impact.
The first question is not “How do we copy the competitor?” It is “What job is the answer trying to complete, and which sources help it complete that job?” The competitor may be named because of third-party consensus, a precise product page, strong local data, original research, reviews, documentation, or simple brand prominence. Each cause suggests a different action.
Define visibility before measuring it
Create separate fields for brand mention, product mention, source citation, owned-domain citation, third-party citation, recommendation position, sentiment, factual accuracy, and referral. A brand can be cited without being named. It can be named without receiving a link. It can be recommended inaccurately. Combining those events into one score obscures the problem.
Semrush’s June 2026 study illustrates the distinction. It logged 3,981 domain appearances for 115 prompts across 14 countries and four answer systems. In that sample, 61.7% were citations without a brand mention, 13.2% combined a citation and mention, and 25.1% were mentions without a citation. The sample is limited and should not become a universal benchmark. It does show why an audit must inspect both the answer text and the source layer.
Build a buyer-task prompt set
Start with customer decisions: understanding a problem, comparing approaches, creating a shortlist, evaluating risk, checking compatibility, estimating effort, finding a local provider, and choosing a next step. Use language from sales calls, site search, support tickets, reviews, paid-search terms, product research, and customer interviews.
Organize prompts into clusters and label their intent. Include non-branded prompts, branded verification prompts, competitor comparisons, use cases, alternatives, and questions that expose objections. Avoid leading prompts such as “Why is our company the best?” They flatter the test instead of measuring discovery.
Prompt register fields
- Prompt text and cluster
- Buyer stage and intended task
- Market, language, and location
- Engine and interface
- Account or anonymous state
- Run date and repetition number
- Brands mentioned and their order
- Sources cited and exact URLs
- Material claims and accuracy notes
- Observed next action
Repeat high-value prompts. BrightLocal’s September 2026 local AI study ran 200,085 non-branded local searches across three platforms and 1,300 business locations, analyzing 1.9 million citations. It found that recommendations changed across repeated runs and locations, although some businesses reappeared more than half the time. That research concerns local queries and vendor-collected data, but it reinforces the need to measure persistence rather than a single answer.
Trace the sources behind each competitor mention
For every competitor appearance, record the cited pages and classify them: competitor-owned page, news or trade publication, review site, directory, community, video, documentation, marketplace, academic source, public record, or another type. Then read the source. Identify the exact passage, fact, comparison, review pattern, or entity relationship that may support the answer.
Some mentions have no visible citation. Search for corroborating sources that describe the competitor in the same way. Look at brand mentions, category associations, anchors, reviews, awards, partnerships, expert biographies, case studies, product documentation, and location listings. The goal is an evidence graph, not a backlink count.
Competitor evidence worksheet
| Observed answer | Supporting source | Evidence type | Owned? | Confidence | Action |
|---|---|---|---|---|---|
| Competitor named for an integration | Current technical documentation | Compatibility | Yes | High | Audit our integration documentation |
| Competitor called a local option | Location pages and directories | Place and availability | Mixed | Medium | Resolve local fact gaps |
| Competitor recommended for expertise | Trade articles and named researchers | Third-party authority | No | Medium | Develop original expert evidence |
This example is hypothetical. Confidence should reflect how directly the source supports the answer, not how attractive the opportunity looks.
Compare page fit, not just domain strength
Open the competitor URL and the best corresponding page on your site. Compare the task each page solves. Does the competitor state the answer earlier? Define the product or service more precisely? Provide current specifications? Show availability? Cite original sources? Explain limitations? Include a comparison table, method, or example? Connect the evidence to a clear entity?
Yext’s July 2026 study of brand-owned AI citations covered 10 businesses and 1,800 United States locations. It found that broad category prompts often cited local pages, while 70% of named intent queries favored intent pages. That narrow dataset does not prove a ranking formula. It offers a useful audit question: does your cited candidate actually match the specific task, or does it mention the topic only in passing?
Audit the brand outside its own site
Ahrefs’ December 2025 study analyzed 75,000 brands and millions of responses. Branded web mentions correlated with AI visibility at roughly 0.66 to 0.71 across the systems studied, while the number of site pages had a weak relationship. Ahrefs states that these are correlations, not proof of causation. Do not respond by purchasing arbitrary mentions. Use the finding to investigate whether credible sources discuss the competitor more often, in more relevant contexts, and with clearer category language.
Map which communities, publishers, directories, partners, conferences, review platforms, videos, and databases matter to the buying decision. Evaluate quality and relevance. A niche standard, current documentation page, or expert association may carry more explanatory value than a high-volume general mention.
Check entity consistency and factual confidence
Answer systems may hesitate when names, products, addresses, leadership, pricing models, or service descriptions conflict. Audit the business’s official pages, profiles, structured data, directories, social accounts, documentation, and major independent sources. Identify stale former names, merged products, duplicate locations, inconsistent categories, and ambiguous author identities.
Create a fact ledger with an owner for each important claim. Stable facts need a canonical source. Volatile facts need a date and update trigger. Regulated claims need an approval path. When the audit finds an inaccurate AI answer, determine whether the web contains conflicting evidence before blaming the model.
Assess technical retrieval
Check whether the strongest candidate pages are crawlable, indexable, canonical, internally linked, rendered without hidden dependencies, and available with successful status codes. Review robots rules for relevant crawlers. Inspect noindex, canonical conflicts, redirect chains, pagination, faceted URLs, duplicate content, and JavaScript-only evidence.
OpenAI’s current publisher documentation states that sites seeking consideration for ChatGPT search summaries and snippets should not block OAI-SearchBot. It distinguishes this search crawler from GPTBot. This is one vendor’s access instruction, not an inclusion guarantee. Record controls by platform and preserve the business’s policy choices.
Find content and evidence gaps
Classify each missing appearance by root cause:
- No suitable page: the site does not answer the task.
- Weak page fit: the answer is buried inside a broad page.
- Insufficient evidence: claims lack methods, sources, examples, or current data.
- Entity conflict: systems cannot confidently connect the brand, product, person, or location.
- Limited third-party support: credible independent sources rarely describe the brand in the relevant context.
- Retrieval barrier: crawlers or users cannot reliably access the page.
- Conversion gap: the cited page answers the question but offers no appropriate next step.
- Sampling noise: the difference does not persist across runs.
Prioritize gaps by buyer value, confidence, feasibility, risk, and reuse. A product-specification correction may be more urgent than a broad thought-leadership article because it affects both answer accuracy and active sales conversations.
Audit negative, missing, and inaccurate visibility
A brand does not win simply because it is mentioned. Record whether the answer assigns the correct category, market, features, audience, price model, and limitations. A prominent inaccurate recommendation can create support burden or regulatory risk. A negative mention may reflect a real product issue, stale third-party information, or unsupported answer synthesis. Each requires a different response.
Create an error register with the claim, engine, prompt, date, cited sources, correct fact, authoritative source, severity, and owner. Correct owned pages and major external profiles first. If a third-party source is wrong, follow its documented correction process. Do not attempt to bury accurate criticism with synthetic content.
Include a control set
Track prompts where the brand already performs well and prompts unlikely to be affected by the planned work. The first group reveals whether improvements elsewhere create regressions. The second helps the team see ordinary answer volatility. Without controls, any market-wide change can look like the effect of the audit.
Keep the core prompt wording stable for trend comparisons, but add exploratory prompts separately as customer language changes. Version the prompt library. If an engine introduces a new interface, record the break rather than presenting the new series as perfectly comparable with the old one.
Connect findings to sales and support evidence
Ask sales teams whether the audited questions match live deals and whether cited competitors actually appear in shortlists. Ask support teams which incorrect expectations cause friction. Referral and conversion data can show which answer surfaces send engaged visitors, although attribution will remain incomplete.
This step prevents the program from chasing visible but low-value prompts. A competitor mention on a broad educational query may matter less than an inaccurate comparison used by buyers near a purchase. Prioritization should combine visibility evidence with commercial reality.
Turn findings into controlled experiments
Change one coherent evidence unit at a time. Improve a comparison page, publish a methodology, clarify a service-location relationship, correct an entity conflict, or build a source-backed answer to a high-value question. Record the before state, change, publication date, crawl state, and repeated post-change observations.
Do not claim causality when several changes launch together or the engine changes during the test. Use language such as “the mention appeared in four of ten post-change runs after being absent in ten baseline runs.” Preserve the sample and limitations.
A 30-day audit sequence
- Days 1 to 5: agree on buyer tasks, markets, engines, competitors, and definitions.
- Days 6 to 10: run and store the baseline with repeat samples.
- Days 11 to 15: trace citations and build competitor evidence graphs.
- Days 16 to 20: compare page fit, entities, retrieval, and third-party context.
- Days 21 to 25: score gaps and draft experiments with owners and acceptance tests.
- Days 26 to 30: validate business priorities, freeze the baseline, and begin implementation.
Large multilingual or multi-location sites may need a longer sample. The sequence matters more than the calendar: define, observe, trace, diagnose, prioritize, then change.
What the final audit should deliver
Require the prompt register, raw answer evidence where permitted, citation inventory, competitor evidence map, entity conflict log, technical findings, page-fit comparisons, prioritized experiments, metric definitions, and limitations. The executive summary should state which gaps are proven, which are plausible, and which need more sampling.
Fuel Online’s AI SEO packages can support the ongoing measurement and implementation layer. If the main gap is missing evidence, pair the audit with expert-led content creation rather than filling the site with generic answers.
Frequently asked questions
How many prompts should an audit use?
There is no universal number. Use enough prompts to cover important buyer tasks, markets, and products, then repeat the highest-value prompts. Document the sample so readers understand what the conclusions represent.
Why does a competitor appear without being cited?
The system may rely on learned associations, sources not displayed in the interface, or retrieved evidence that names the competitor indirectly. Audit corroborating web mentions and repeat the prompt before drawing a conclusion.
Should we copy pages that receive citations?
No. Identify the task and evidence the page supplies, then produce a better truthful resource from your own expertise. Copying structure or language can create duplication and still leave the underlying evidence gap.
Can an AI visibility audit guarantee future mentions?
No. It can produce a defensible baseline, diagnose observable gaps, and guide experiments. Engines, sources, and answers change outside the auditor’s control.
How often should the audit be repeated?
Monitor priority prompts on a regular cadence and rerun the deeper audit after significant product, market, website, or platform changes. Preserve the same core sample while adding new buyer questions deliberately.
The useful outcome is a better evidence system
A good AI visibility audit does more than count competitor mentions. It shows which questions matter, which sources shape the answers, where the business lacks clarity or proof, and which controlled changes are worth making. Even when an engine changes, the resulting pages, facts, research, and measurement discipline continue to help buyers.
