An AI citation audit checks whether AI tools mention or link to your business when buyers ask questions in your category, and which sources they rely on instead. You build a fixed set of real buyer questions, run them through ChatGPT, Perplexity, Gemini and Google’s AI Overviews and AI Mode, log what each answer says and cites, and turn the gaps into a plan. Done monthly, it becomes the most honest picture you have of your AI visibility.
This checklist works for agencies running audits for clients (it’s easy to white-label) and for businesses doing it themselves.
Key takeaways
– AI answers cite a lot of sources. Moz found 96% of Google AI Mode responses included at least one citation, with most pulling from 10 or more URLs.
– Citation patterns shift without warning. In Semrush’s 13-week study of 230,000 prompts, ChatGPT sharply cut citations to Reddit and Wikipedia in mid-September 2025, though both stayed among its most-cited domains.
– Community and video platforms dominate many answers. Ahrefs’ September 2026 data put YouTube at 22.9% and Reddit at 18.5% of citation share among the top 50 domains in AI Overviews.
– Use the same prompts every month and log results in a consistent format, so you can see trends rather than one-off screenshots.
– The output of an audit is a list of actions, not a score.
What does an AI citation audit measure?
Four things, for each question in your set:
- Mention: Does the answer name your business?
- Citation: Does it link to your website or a page about you?
- Accuracy: Is what it says about you correct?
- Sources: Which websites does it cite or rely on, and which competitors does it name?
Together, these show whether you’re visible, whether you’re represented correctly and where the AI is getting its information.
Step 1: Build the prompt set
Start with 20–50 questions your buyers actually ask. Pull them from:
- Sales call notes and common objections.
- Support tickets and onboarding questions.
- Search Console queries for your key pages.
- “People also ask” boxes in Google.
- Review sites and community forums in your industry.
Cover the full buying journey:
| Stage | Example prompt format |
|---|---|
| Problem | “Why does [problem] happen?” |
| Options | “What are the best ways to fix [problem]?” |
| Comparison | “[Option A] vs. [Option B] for [use case]?” |
| Vendor | “Best [service] companies in [city/industry]?” |
| Validation | “Is [your company] good for [use case]?” |
Write prompts the way a real person would, not as keywords.
Step 2: Choose the platforms and settings
At minimum:
- ChatGPT (with search enabled)
- Perplexity
- Google Gemini
- Google AI Overviews (regular Google searches that trigger them)
- Google AI Mode
Optional: Microsoft Copilot and Claude, if your buyers use them.
Keep conditions consistent. Use a logged-out or fresh session where possible, note the location setting, and record the date. AI answers vary between runs, so for your most important prompts, run each two or three times.
Step 3: Log results in a standard format
A simple spreadsheet works. One row per prompt per platform:
| Column | What to record |
|---|---|
| Date | When you ran it |
| Platform | ChatGPT, Perplexity, etc. |
| Prompt | Exact wording |
| Mentioned? | Yes / No |
| Linked? | Yes / No, and which URL |
| Accurate? | Yes / Partly / No, with a note |
| Competitors named | List |
| Sources cited | List of domains or URLs |
| Notes | Anything unusual |
Save screenshots or exports for your most important prompts. They’re useful when explaining findings to a client or your team.
Step 4: Analyze the patterns
After the first full run, look for:
- Your mention rate: in what share of answers are you named?
- Competitor mention rate: who shows up most, and for which stages?
- Top cited sources: which publications, directories, forums, videos or review sites appear repeatedly?
- Accuracy problems: wrong services, outdated pricing, old addresses, confusion with similar names.
- Stage gaps: maybe you appear in vendor questions but never in problem or comparison questions.
The source list is often the most valuable output. Those are the sites AI tools trust in your category, and the places you need to be.
Step 5: Turn findings into actions
Each finding should map to a fix:
| Finding | Typical action |
|---|---|
| Not mentioned for key questions | Create or upgrade a page that answers that question directly, with proof |
| Inaccurate information | Correct the source: your site, Google Business Profile, directories, LinkedIn |
| Competitor cited from a third-party list | Pitch or apply to that list, review site or publication |
| Video sources dominate | Create a short, well-titled, captioned video answering the question |
| Forum threads dominate | Monitor and participate genuinely where appropriate; don’t spam |
| Only cited for brand questions | Build content for problem and comparison stages |
Step 6: Repeat monthly and report trends
Re-run the same prompts monthly. Report mention rate, citation rate, accuracy issues fixed and new sources gained. Quarterly, review the prompt set and add new questions as your offer or market changes.
Semrush’s finding that ChatGPT’s citation patterns shifted sharply in a single week is a good reminder: one month’s results aren’t permanent. Trends over several months matter more than any single snapshot.
How does this connect to Search Console?
For Google’s AI features specifically, Search Console’s Generative AI performance report shows how often your pages appear in AI Overviews and AI Mode, by page. It doesn’t show queries or competitors, so the manual audit fills that gap. Use both: the report for scale, the audit for context.
Illustrative example: a first audit summary
Illustrative example (placeholders, not client data): A [regional accounting firm] runs [X] prompts across five platforms. It’s mentioned in [X] of [X] answers, mostly for “[firm name] reviews” questions. For “best small business accountant in [city],” AI tools name three competitors and cite a local business journal list and a review site the firm isn’t on. Two answers list a service the firm no longer offers. Actions: update the services page and directory listings, apply to the review site, pitch the business journal, and build a page answering “How much does a small business accountant cost in [city]?”
How should agencies white-label this audit?
- Standardize the template so every client gets the same structure.
- Agree the prompt set with the client. Their sales team knows the real questions.
- Lead with the action list, not the raw data.
- Include a one-page summary for executives: mention rate, top gaps, top three actions.
- Offer it as a recurring service, monthly or quarterly, rather than a one-off.
What should businesses do if they’re running it themselves?
Start small: 20 prompts, three platforms, once a month. Put it on someone’s calendar. The value comes from consistency over time, not from a perfect first run.
How do you score accuracy fairly?
Accuracy is where audits get subjective, so agree a simple rubric before you start:
- Accurate: everything the answer says about the business is correct and current.
- Partly accurate: mostly right, but with an outdated detail (old hours, a former service, a previous address) or a vague description.
- Inaccurate: a material error, such as wrong services, wrong location, confusion with another company or a false claim.
For each “partly” or “inaccurate” result, note the specific error and where it likely came from. Many errors trace back to an old directory listing, an outdated About page or a third-party article. Fixing the source is more durable than hoping the AI changes on its own.
What should the one-page summary include?
Executives won’t read the spreadsheet. Give them one page:
Illustrative summary (placeholders, not client data):
– Mention rate: named in [X]% of answers ([up/down] from [X]% last month).
– Link rate: linked in [X]% of answers.
– Accuracy: [X] errors found; [X] fixed since last audit.
– Top competitors in answers: [Competitor A], [Competitor B], [Competitor C].
– Most-cited sources we’re missing from: [Site 1], [Site 2], [Site 3].
– Top three actions this month: [Action 1], [Action 2], [Action 3].
That fits on a slide and leads straight into decisions.
How much time does an audit take?
A first audit with 30 prompts across five platforms typically takes most of a day, including setup, running the prompts and writing up findings. Monthly re-runs are faster once the template exists, often two to three hours. Tools that automate prompt tracking can cut the running time, but someone still needs to read the answers, judge accuracy and decide what to do. That judgment is the part clients actually pay for.
What are the common mistakes?
- Running prompts once and drawing big conclusions. Answers vary.
- Using keyword-style prompts. Write like a buyer.
- Ignoring accuracy. Being mentioned incorrectly can be worse than not being mentioned.
- Collecting data with no actions. Every finding needs a next step.
- Changing the prompt set every month. You lose the ability to compare.
Frequently asked questions
What is an AI citation audit?
It’s a structured check of whether AI tools like ChatGPT, Perplexity, Gemini and Google’s AI features mention or link to your business for buyer questions, and which sources and competitors they rely on instead.
How many prompts should an AI citation audit include?
Usually 20–50 real buyer questions covering the full buying journey. Start smaller if you’re doing it yourself and expand over time.
How often should I run an AI citation audit?
Monthly is a good rhythm for tracking trends, with a quarterly review of the prompt set. AI answers change often, so single snapshots can mislead.
Can I automate AI citation tracking?
Several tools offer automated prompt tracking. They save time, but review results manually for accuracy and context, especially for your most important prompts.
What should I do if AI tools say something wrong about my business?
Fix the information at its likely source: your website, Google Business Profile, directories, LinkedIn and review sites. Then recheck in the next audit.
Which AI platforms should I audit?
At minimum, ChatGPT, Perplexity, Gemini, Google AI Overviews and Google AI Mode. Add Copilot or Claude if your buyers use them.
Want an AI citation audit done for you?
We run AI citation audits for businesses and white-label them for agency partners, from prompt sets to source analysis and a clear action plan.
Prefer to talk? Call (813) 397-3665 or email strategy@seooutsourcing.com.
