Audit the buyer questions, answer evidence, recommendations, competitors, and citations together.
To perform an AI visibility audit, define a commercially meaningful question set, record the conditions and raw answers, distinguish mentions from recommendations, inspect competitor displacement and citations, prioritize only supported evidence gaps, and retest a stable core. The result is a reviewable baseline—not a universal AI ranking or a promise that one content change caused an answer to move.
Where does recommendation evidence break down?
A B2B SaaS AI visibility audit should not begin with “How visible are we?” That question is too broad to drive a decision. Begin with a defined buyer, a commercial situation, and the questions that could place a vendor into—or remove it from—a considered set.
The objective is to identify which questions recommend competitors, which sources support those answers, and which evidence gap is both credible and valuable enough to address first. Traffic, mentions, and a composite score can provide context, but they do not replace that evidence chain.
How to perform an AI visibility audit for your brand, from category definition to retest.
- Step 1
Define your brand and product category
Record the exact brand, product category, target audience, country, and language the audit should represent.
- Step 2
Identify relevant competitors
Choose the real alternatives that could plausibly enter the same buyer shortlist; keep this comparison set explicit.
- Step 3
Build branded questions
Test recognition and factual accuracy with questions that name the brand, product, pricing, integrations, or other validation needs.
- Step 4
Build unbranded commercial questions
Select discovery, comparison, and buyer-fit questions where the brand must earn consideration without being supplied in the prompt.
- Step 5
Run questions consistently
Attach the provider, model or interface, country, language, date, wording, and repetition number to each retained answer.
- Step 6
Classify mentions
Record whether the brand is present or absent without treating a passing mention as an endorsement.
- Step 7
Classify recommendations
Mark whether the answer presents the brand as a plausible choice for the stated need and retain the supporting language.
- Step 8
Check competitor displacement
Record displacement only when the target is not recommended and a relevant predefined competitor is recommended in its place.
- Step 9
Analyze citations
Record the sources displayed with the answer and note which claims, products, or competitors each source appears to support.
- Step 10
Separate owned and third-party sources
Distinguish company-controlled product, pricing, documentation, and proof pages from independent evidence.
- Step 11
Form evidence-gap hypotheses
Look for missing product facts, weak category language, unclear positioning, technical access issues, or limited independent proof without claiming causation.
- Step 12
Prioritize an action
Assign the highest-value supported action to an owner with acceptance criteria, confidence, and a review date.
- Step 13
Retest a stable core
Repeat the commercially important questions under recorded conditions and compare like with like while acknowledging model variability.
Build the set around buying decisions, not prompt volume.
Discovery
Questions that ask for a category or solution for a defined team, workflow, or constraint.
Comparison
Alternatives, shortlists, migrations, integrations, and trade-offs between credible vendors.
Validation
Pricing approach, security, implementation, proof, fit, and other facts needed before a buyer advances.
Avoid loading the set with branded questions you already win, prompts that no plausible buyer would ask, or tiny wording variations counted as independent demand. Record why each question matters and which decision it represents. That rationale is what lets a team remove a weak question instead of protecting a flattering metric.
Treat presence, endorsement, and proof as separate observations.
A mention means the answer names the company. A recommendation means it presents the company as a plausible choice for the stated need. A citation identifies evidence used or surfaced by the answer. A vendor can be mentioned but not recommended, recommended without an owned citation, or absent while a competitor is supported by several authoritative third-party sources. Those are different problems and should not collapse into one score.
For each competitor advantage, inspect the cited page and the claim it supports. The gap may be technical accessibility, unclear positioning, missing product detail, weak proof, inconsistent pricing information, or insufficient independent evidence. The proposed action should name the observed gap—not assume that publishing another article will fix it.
Keep conditions and interpretation attached to every finding.
- Buyer context
- Persona, company stage, category, use case, constraints, and commercial intent.
- Question
- The exact wording tested, kept stable enough to retest and specific enough to reveal a decision.
- Run conditions
- Provider, model, country, language, date, and repetition number.
- Observed answer
- Raw response, mention, recommendation, position, citations, competitors, and factual accuracy.
- Interpretation
- Evidence gap, confidence, proposed action, owner, acceptance criteria, and retest date.
Turn one observed omission into a testable next step.
Fictional brand—not a customer result
- Brand
- AcmeCRM
- Question
- “What is the best CRM for small SaaS teams?”
- Observation
- AcmeCRM is absent; Competitor A is recommended; a third-party page about Competitor B is cited; AcmeCRM's owned content is not referenced.
A defensible interpretation
The observation establishes displacement for this question under these run conditions. It does not establish why the model produced the answer. The next review should compare AcmeCRM's category and audience language, product evidence, pricing clarity, technical accessibility, and independent proof with the evidence supporting the recommended competitors.
Any resulting change should have an owner and acceptance criteria, then be evaluated by retesting the same core question set. A changed answer is a new observation, not automatic proof that one edit caused it.
A spreadsheet can establish the method; a tool can make the evidence easier to repeat.
Manual AI visibility audit
Useful for learning the classifications and testing a small set cheaply. It gives the researcher direct context, but copying answers, normalizing run conditions, retaining citations, and comparing repetitions becomes time-consuming and error-prone as the set grows.
Tool-assisted AI visibility audit
Useful when the team needs consistent collection, attached evidence, repeatable classifications, or a larger approved set. Tool scores still require documented definitions and judgment. RecoProof's free check is a bounded automated Sonar Pro snapshot; its paid audit expands provider coverage and adds human review rather than claiming universal interface coverage.
One run is evidence, not a permanent ranking.
Answers can change with model versions, retrieval, location, date, phrasing, and product conditions. API output may differ from a personalized consumer session. A defensible report states those boundaries, retains raw answers, and avoids extrapolating beyond the tested question set.
Retest the decision, not just the wording.
Keep a stable core question set, preserve run conditions, repeat uncertain observations, and define acceptance criteria before making a change. Monitoring is valuable when it shows whether a corrected evidence gap persists across time—not when it produces more charts.
Don't want to run the audit manually?
Start with a bounded free snapshot of five commercial questions. RecoProof keeps the observed answers, recommendations, competitors, citations, and limitations reviewable so you can decide whether deeper investigation is warranted.
Run the bounded free auditUse the evidence next
Review the report, assign the highest-value supported action, define acceptance criteria, and retest under recorded conditions.
Turn the evidence into a content and authority plan Choose the AI visibility metrics to track