How to Run an AI Visibility Audit for Your Chennai Business (Step by Step)

What an AI visibility audit is, and what it is not

An AI visibility audit answers four questions about your business:

1. Presence. When people ask the questions your customers ask, how often are you named?
2. Accuracy. When you are named, is the description right: services, location, prices, reputation?
3. Sources. Which pages and sites do the tools rely on, and are any of them yours?
4. Causes. What technical, content or reputation gaps explain the pattern?

It is not a ranking report. There is no stable “position one” in a generated answer, and it is not a one-off screenshot. It is a repeatable measurement. We use the same approach in our own AI visibility case study, where we tracked 30 queries across four platforms over 24 weeks. For the bigger picture of what you are trying to improve, read AI visibility for Chennai businesses.

Why a single screenshot misleads

In January 2026, Rand Fishkin of SparkToro and Patrick O’Donnell of Gumshoe.ai published a study of 2,961 prompts across ChatGPT, Claude and Google’s AI features. They ran 12 brand-recommendation prompts 60 to 100 times each on every platform. The headline findings:

• ChatGPT and Google’s AI features returned the same list of brands less than 1% of the time across repeated runs of the same prompt.
• The same list in the same order appeared less than 0.1% of the time.
• Nearly every response differed in the brands listed, their order or the number of items.
• Despite that noise, some brands appeared again and again. In one set of headphone prompts, four brands appeared in 55% to 77% of responses.

The authors concluded that position in a list is an unreliable measure and that the share of runs in which a brand appears is more stable. They also noted limits: the study was done in partnership with a company that sells AI tracking tools, it is not peer-reviewed, and it is unclear how many runs are enough. It is still the best public evidence we have for why you should never base a decision on one result.

The practical lesson is to measure frequency. Ask more than once, and report “named in 6 of 10 runs”, not “we are second”.

Step 1: Build a list of 25 to 30 real questions

Use the words customers use. Pull them from sales calls, WhatsApp enquiries, Search Console and reviews. Cover six types so that you see the whole customer journey.

Include a few questions in Tamil or mixed Tamil and English if your customers use them. Keep the list fixed so that months can be compared. If you change a question, log it as a new one.

Step 2: Run the questions the same way every time

• Platforms. At minimum ChatGPT with search on, Perplexity, Gemini and Google (for AI Overviews and AI Mode). Add others your customers use.
• Repeat each question. Run each one at least three times per platform, ideally five. More runs give steadier percentages. Our three to five is a practical compromise, not a figure from the study.
• Same conditions. Use a fresh chat or a private window, the same wording, the same device and similar time of day. If you are signed in, past chats can colour the answer, so note it.
• Same location. Local answers can depend on location, so run from Chennai or state the location in the question.
• Date everything. Systems change quickly. A dated log is the only way to see real movement.

Thirty questions, four platforms and three runs is 360 answers. It is a lot, but it takes a few focused hours, and you can split it across a week.

Step 3: Record what you see

A spreadsheet is enough. One row per answer, with these columns:

Step 4: Turn the log into a few numbers

Keep the scoring simple enough to explain to a colleague.

• Visibility percentage. Answers in which you are named, divided by total answers, for the whole set and for each question type and platform.
• Citation rate. Answers that link to or cite your site, divided by total answers.
• Accuracy rate. Answers about you that are correct, divided by answers that mention you.
• Share of voice. Your mentions compared with each competitor across the same set.

Look at the pattern, not just the totals. A business that is strong on brand questions but absent from category questions has a discovery problem. One that is present but inaccurate has an information problem. The fixes are different.

Step 5: Study the sources the tools use

This step often tells you the most. List every site the answers cite and count how often each appears. You will usually find a small set of directories, publications, forums and competitor pages that do most of the work.

• If a directory is cited often and you are missing or wrong there, fix it. See business listing sites in Chennai.
• If a publication is cited often for your category, ask how to be included, as discussed in our link building guide.
• If competitors are cited from their own pages, study those pages: how they answer, what they prove, how current they are.
• If your own pages are cited, note which, and build more like them.

Step 6: Check technical access

Sometimes the cause is mundane: the tool cannot reach you.

1. Robots rules. OpenAI documents OAI-SearchBot for ChatGPT search results, GPTBot for training and ChatGPT-User for user-initiated visits. Blocking OAI-SearchBot means your pages may not appear in ChatGPT search answers. OpenAI says robots.txt changes can take about 24 hours to be reflected.
2. Host and security layers. A firewall or bot-protection rule can block legitimate crawlers even when robots.txt allows them. Check your server logs.
3. Google eligibility. To be a supporting link in Google’s AI features, a page must be indexed and eligible to show with a snippet. Check indexing in Search Console and review any nosnippet or max-snippet rules.
4. Speed and rendering. Important text should be in the HTML, not hidden behind scripts. See the technical SEO checklist for Chennai sites.
5. Search Console. Google reports traffic from AI features within the Web search type, so you can track clicks there.

Step 7: Diagnose and prioritise fixes

Pick the two or three fixes with the largest gap and the least effort, do those first and retest. Do not try to fix everything at once, because then you will not know what worked.

A sample result, so you know what to expect

The numbers below are invented to show the format. They are not from a real client.

In this fictional business, people who already know the name get a good result, but discovery is poor and the few price answers are wrong. The plan would start with a pricing page, a corrected profile and answer-led content for problem questions.

Common audit mistakes to avoid

• Testing only your brand name. Almost everyone looks good when asked about themselves. Discovery questions are where the real gaps show.
• Using leading prompts. “Why is [my business] the best dentist in Adyar” invites a flattering answer. Ask neutral questions the way a stranger would.
• Judging from one platform. A good result on one tool says little about the others, because they use different sources.
• Ignoring wrong answers. An inaccurate price or address costs enquiries. Log every error and trace its source.
• Changing the question list each month. You lose the ability to compare. Add new questions as new rows, and keep the old ones.
• Treating the audit as the goal. The audit is a diagnosis. Its value is the fixes you make and the retest that follows.

If you only have an hour, test fifteen category and problem questions once on each platform and write down who gets named. Even that rough picture is usually enough to show whether you have a real gap.

Manual audit or a paid tool?

Start manually. It teaches you what the tools say and what your customers see, and it costs only time. Paid trackers can run far more prompts and save effort, but treat them with care. Ask any vendor how many times each prompt is run, how location and language are handled, how they deal with answer variation and what the percentages actually mean. Fishkin’s own advice was to ask providers to show their method. A tracker that reports a single rank for a generated answer deserves scepticism.

How often to repeat the audit

Monthly is a good rhythm, with a quick check after any major change to your site or profiles. Quarterly is acceptable for small businesses with limited time. Compare like with like, and judge the trend over several months, not a month-to-month wobble. If you want an experienced team to run it, see our AI SEO agency page or how to choose an AI SEO agency in Chennai for what to ask before you hire anyone.