How to Run Your First AI Visibility Audit
Step-by-step guide to audit AI answers: build prompt sets, test ChatGPT/Perplexity/Google, score results, and prioritize fixes.
GTM Engineer at AskWatch.ai
8 min read
On this page
I’d start an AI visibility audit with 20-30 customer questions, fixed test settings, and a scorecard. The goal: check whether AI answers name your brand, link to your site, describe it correctly, or recommend competitors.
Here’s the process I’d follow:
- Build the prompt set. Use actual customer questions, and keep branded questions separate from discovery searches.
- Test and record. Run the same questions across ChatGPT, Perplexity, and Google AI Overviews. Save answers and source links, and repeat priority questions at least 3 times on separate days.
- Choose the next fixes. Review recurring gaps, correct wrong facts and access issues first, then rank your top 5 actions.
I’d save that baseline, check unstable prompts weekly, and repeat the full audit monthly. The key is <u>using the same rules each time</u> - and treating AI visibility as a separate measure from traffic and leads. This is because AI Overviews differ from traditional search in how they cite sources and impact traffic.
How to Audit Your Business for AI Visibility Step-by-Step Guide
Before you begin your manual audit, you can use our AI Overview checker to see which of your pages are currently cited in Google.
Step 1: Build Your Customer Prompt Set
Your prompt set is the baseline for every platform test.
Start with 20-30 prompts for a first manual audit. Use 30-100 only for broader automated coverage. Pull questions word for word from sales calls, support tickets, and customer research. Use Google Search Console to find unbranded queries already driving traffic or interest. Then check Reddit, YouTube, and G2 for comparison, alternative, and purchase questions.
Cover Discovery, Evaluation, and Purchase Questions
Include discovery, problem-first, comparison, alternative, and purchase-stage questions. Cover lead routing, team size, pricing, and reliability.
Keep a smaller branded set to check product, pricing, and positioning accuracy. Don’t make branded questions your primary set.
Record every prompt exactly as written before testing.
Create a Fixed Prompt Library
Build a spreadsheet with each prompt’s ID, verbatim text, intent band, customer intent, geographic modifier, priority, and business owner.
Prompt ID Exact wording Intent band Customer intent Geographic modifier Priority Business owner P01 How to route leads to a 5-person sales team? Problem-first Discovery: solve lead routing None - - P02 Best CRM for a team of five under $100 a month. Constrained comparison Evaluation: shortlist within budget None - - P03 What should I use instead of Salesforce? Alternatives Evaluation: find a replacement None - -
Remove near-duplicates that express the same need. Keep differences in budget, use case, or location when they change what the customer needs. Give priority to questions close to a buying decision. Draw branded and purchase-stage questions from your own customer evidence - don’t invent wording just to fill a category.
Freeze the library once testing starts. Keep wording, spelling, location, and intent identical across platforms. Save edits as new versions rather than overwriting the baseline. Test old and new prompts in the same audit cycle so you can separate wording changes from visibility changes.
Step 2: Test Platforms and Record Answers
Standardize Tests and Repeat Priority Prompts
With your prompt library frozen, run each prompt the same way across ChatGPT, Perplexity, and Google AI Overviews during a fixed audit window. Record brand mentions, citations, and competitor appearances. Keep Google AI Overviews and Google AI Mode separate.
Run each priority prompt at least 3 times on separate days to account for variation in answers. Log test errors as their own outcome - not as missing brand mentions.
Log Mentions, Citations, and Recommendations
Use one scorecard row per prompt-platform run so you can compare answers against a baseline.
Attach the full answer, screenshot, and source URLs to each run ID. Check for brand mentions in the answer text, not just in citations.
| Scorecard fields | What to record |
|---|---|
| Run ID / Number | A distinct identifier for each run |
| Prompt ID | Connects different runs of the same prompt |
| Exact prompt | The question, word for word |
| Platform / search mode | The platform and search mode tested |
| Run settings | Model, location, account state, and search settings |
| Answer availability | Shown, Unavailable, or Error; keep an unavailable overview separate from an absent brand |
| Brand mentioned | Yes or no |
| Brand position | Numerical rank or null if not mentioned |
| How the brand is described | How the answer describes or recommends your brand |
| Domain cited | Whether the answer links to your owned site |
| Cited URLs | The exact source URLs used by the AI |
| Competitors named | All rival brands named in the same answer |
| Answer text | The full response for verification |
| Evidence notes | Links to the full answer and screenshot |
Document any differences between automated and manual checks, including location, account state, model, and search mode.
Track Repeat Audits With AskWatch
After logging the first pass, use AskWatch to repeat audits with the same prompt library, test settings, and scoring rules.
Use these logs to identify the biggest visibility gaps in Step 3.
Step 3: Find and Prioritize Visibility Gaps
Use the scorecard to group runs by prompt, platform, and gap type before ranking gaps. Prioritize recurring gaps; ignore isolated outliers.
Review Cited Sources
Open pages that appear repeatedly in citations. Check each for currency, accuracy, relevance, and tone against current product documentation. Review competitor pages, comparison pages, reviews, and industry publications to see which evidence the AI uses, what it misses, and what your brand needs to fix or build.
Use this worksheet to log only pages the AI cited:
| Cited domain | Cited page | Topic | Brand or competitor reference | Factual claim supported | Missing evidence or coverage |
|---|---|---|---|---|---|
| Your owned domain | Exact cited URL | Product, pricing, or use case | Record the named brand | Check against current documentation | Correct stale facts or missing details |
| Competitor’s domain | Exact cited documentation or comparison URL | Feature or buying decision | Record the named competitor | Identify the supporting passage | Check whether equivalent evidence is missing |
| Third-party domain | Exact cited review or article URL | Reviews or category recommendations | Record which brands appear | Note the claim and its tone | Check listing accuracy or editorial coverage |
Turn each recurring citation pattern into a fix, a content gap, or an outreach target.
Group Actions: Fix, Build, and Influence
Fix misleading descriptions, outdated pricing, wrong URLs, and access blockers first. Verify bot access before changing robots.txt or firewall rules. Make company and product facts consistent across your website and relevant third-party profiles.
Build product, use-case, pricing, or comparison content when high-value prompts repeatedly lack brand coverage or supporting evidence. Answer the buying question first, then back it up with specifics.
Influence recurring third-party coverage through accurate listings, editorial outreach, and community participation.
With the gaps grouped, rank those that affect visibility most.
Rank Your Top 5 Visibility Gaps
Rank gaps that recur across runs and platforms. Pay particular attention to technical blocks, factual errors, missing third-party coverage, purchase-stage misses, and awareness gaps. Address serious factual errors and access blockers before lower-risk content work.
Give each action an owner, supporting run IDs and URLs, an expected outcome, and a due date. Link each ranked item to its cited URLs so you can trace it back to the baseline. Define the outcome as a correction or deliverable you can verify, then check for answer changes in the next audit.
Conclusion: Save Your Baseline and Plan the Next Audit
Before you save the report, confirm the audit scope, frozen prompts, test settings, and ranked gaps. Archive the full answers and run logs with it. This record becomes the baseline for your next audit.
Date the baseline and include your four core metrics. Record each metric’s denominator and specify which answers counted toward accuracy. These metrics summarize the prompt set, platform tests, and gap analysis you’ve completed. Include your strongest platforms, weakest prompt categories, and top 5 next actions.
Repeat the Audit With the Same Rules
Rerun the same test set on a fixed schedule. Review your most unstable priority prompts weekly using a ChatGPT visibility tracker and rerun the full baseline monthly. Keep the prompts, platforms, session settings, run counts, and scoring rules unchanged. If you revise the prompt library, run the old and new sets together to keep the results comparable.
Treat the baseline as a measurement record, not a performance verdict. Look beyond the overall score: compare mention and citation rates, recommendation position, source coverage, and competitor presence. Track visibility separately from traffic and leads. If results improve after a content update, review the change - but don’t assume the update caused it.
FAQs
How do I calculate my AI visibility metrics?
Set a baseline by testing the same set of buyer-intent prompts across your target AI engines. Run each prompt at least three times. Use a spreadsheet or AskWatch to record brand mentions, position, competitors, and cited domains.
Calculate share of answers by dividing brand mentions by total answers. Break down the results by engine and intent. Track citations separately from mentions so you don’t overstate visibility, and review the results monthly.
What counts as a meaningful visibility improvement?
Better visibility means more than getting your brand named. Across repeated tests using the same set of real buyer prompts, your brand should appear more often (a higher “share of answers”), rank better on average when mentioned, and be cited as a source more often.
Those gains should exceed normal run-to-run variation. Run each prompt multiple times - for example, at least three runs per prompt - and track trends over time instead of relying on a single check.
How soon should I expect fixes to affect AI answers?
Your fixes may start appearing in AI citations much sooner or much later. Google AI Overviews and AI Mode can use fresh web sources, and ChatGPT Search can too. Schema updates, entity work, and third-party mentions still need time to be crawled and indexed, and some answers depend on retrieval systems or later model updates.
Plan for a long-term process, not an overnight change. Use a tool like AskWatch to keep a consistent measurement baseline, and check results through monthly audits rather than expecting an immediate shift.
