AI Visibility Audit: See Exactly Where AI Mentions You

2. What an AI Visibility Audit Actually Measures

Quick Answer: An AI visibility audit runs a fixed set of real buying questions across the major answer engines. It logs whether your business is recommended, cited, named or absent in each reply, records which sources the engine drew from instead, and dates the whole sample so it can be repeated.

Four things get recorded, and all four matter separately. Being named is not the same as being linked, and being linked is not the same as being recommended.

• The prompt set. The exact wording of every question, so the next run is comparable rather than approximate.
• Your position in each answer. Recommended, cited with a link, mentioned in passing, or missing entirely.
• Who appeared instead. The competitors and source pages the engine pulled from — often review roundups and forums rather than company websites.
• The date. Answers move. An undated screenshot proves nothing a month later.

That last point separates an audit from a demo. Anyone can show you one unflattering ChatGPT reply. A baseline is a dated, repeatable sample you can hold the next report against — the same discipline behind measuring your ChatGPT visibility over time and benchmarking AI search share of voice against rivals.

3. Why Your Analytics Cannot Answer This Question

Quick Answer: Google folds AI Overviews and AI Mode into ordinary web search reporting rather than breaking them out, and ChatGPT, Gemini, Perplexity and Copilot give site owners no reporting at all. So no dashboard you already own can tell you whether an answer engine named you.

Google’s own documentation states that sites appearing in AI features are included in overall search traffic in Search Console, inside the Web search type. They are counted, but not separated. You can see clicks fall. You cannot see the answer that replaced them.

The chat engines are blunter still. No property to verify, no impressions report, no query list. The only way to know what ChatGPT says about you is to ask it and write down the answer — which makes this a sampling exercise, not a data pull. Referral-side tracking helps a little, and setting it up is covered in tracking AI search traffic in GA4, but it only catches people who clicked. Most never do.

4. The Four Levels of Visibility Worth Recording

Quick Answer: Score every prompt on one of four levels — absent, mentioned, cited or recommended. Across tracked Malaysian SME buying prompts, roughly half sit at absent and only a small share reach recommended, which is why a single pass or fail number hides the real gap.

Collapsing this into one percentage is the most common reporting mistake. A business named in eight answers but recommended in none has a different problem from one nobody sees at all.

The right-hand column is the useful part. Each level fails for a different reason, so each has a different first move — page coverage, then identity, then formatting, then outside corroboration. The identity layer is unpacked in teaching Google and AI who your company is.

5. Building the Prompt Set: The Step Most Audits Skip

Quick Answer: Build 20 to 40 prompts in the wording a real buyer would use, covering problem, comparison, price and location questions. Weak audits reuse keyword lists instead, which produces answers nobody was ever going to ask for.

Keywords and prompts are different objects. Nobody types “seo agency malaysia” into ChatGPT. They type a sentence with a constraint in it. Build the set in this order:

1. Pull the real questions from sales. Ask whoever answers enquiries for the twenty questions they hear most, in the customer’s own words.
2. Add the constraint layer. Budget, location, industry, timeline — “in Penang”, “under RM 5,000 a month”, “for a small clinic”. Constraints are what make an engine name specific companies.
3. Cover all four intents. Problem, comparison, price, and “who does this near me”. Most audits over-index on comparison and miss price entirely.
4. Include the prompts you expect to lose. Sampling only flattering questions produces a sales tool, not a measurement.
5. Freeze the wording. Save the exact strings. Changing one word between runs makes the comparison meaningless.

Twenty to forty prompts is the practical band for an SME. Fewer and the run is noise; more and nobody repeats it monthly. Which questions carry commercial weight is covered further in the 2026 AI search playbook.

6. Why a Single Run Is Not a Baseline

Quick Answer: Answer engines are not stable. Run the same prompt set twice a week apart and a large share of replies name a different set of companies, with no change to any website in between. A credible audit samples repeatedly and reports a range.

This finding surprises most owners, and it is why one dramatic screenshot should never convince you.

Two things follow. Ask any provider how many runs sit behind their numbers, and judge movement over months rather than weeks, because week-to-week wobble swamps any real gain. That is also why a monthly AI SEO retainer should include re-sampling rather than treating the baseline as a one-off.

7. What a Real Audit Report Contains

Quick Answer: A usable report gives you the frozen prompt list and per-prompt scoring across every engine sampled. It also names the companies and source pages that appeared instead, the pages on your site that failed to qualify, and a prioritised fix list with owners.

Ask for a redacted sample before signing anything. These sections should exist:

• The prompt list, in full. Not a summary. You should be able to re-run it yourself.
• Per-prompt, per-engine scoring. One row per prompt, one column per engine, scored on the four levels.
• The replacement set. Who got named instead, and which pages the engine cited — usually directories, forums and roundups.
• Page-level diagnosis. Which of your pages should have qualified for each prompt, and what disqualified them.
• A prioritised fix list. Ordered by effort against likely movement, with a named owner per item.
• The re-sample date. A stated next run, otherwise nothing gets measured.

If the deliverable is a slide deck of screenshots and a proposal, you bought a pitch. This is the evidence layer sitting underneath what an AI search agency actually does and what generative engine optimisation services include.

8. What Malaysian Baselines Usually Show

Quick Answer: Absence rates vary sharply by sector. Education and professional services baseline strongest because their buyers ask written questions that get answered in written content. F&B and B2B supply baseline weakest, since their visibility lives in maps, menus and catalogues rather than in text an engine can quote.

Knowing your sector’s usual starting point stops you reading a poor first result as a crisis, or a decent one as a finished job.

The pattern is consistent: sectors that publish written answers get quoted, and sectors relying on images, listings or spec sheets do not. For local service businesses, much of the early gain comes from listings hygiene, which is why professional Google Business Profile management often moves the needle before any content work starts.

9. What Moves After the Audit, and When

Quick Answer: Expect month one to be flat, because it is the baseline. Named-in-answer rates typically begin moving from month two once restructured pages are recrawled, and citation rates follow a month behind that. Six months is a fair first judgement point.

Setting this expectation before work starts is what stops a sensible programme being cancelled in week six.

Two habits keep this honest: re-run the frozen prompt set on the same day each month, and never edit the prompts to chase a better number. Where the gap is writing, content SEO services carry the load; where it is credibility, corroboration such as structured entries on Wikidata and a Google Knowledge Panel does more.

10. What an AI Visibility Audit Should Cost

Quick Answer: In Malaysia, a standalone audit for an SME generally runs between RM 2,000 and RM 6,000 as a one-off, and is often waived or credited when it opens an ongoing retainer. Free instant checkers are lead magnets, not audits.

Price tracks the number of prompts, engines and runs, and whether a human reads the answers.

• Free instant checkers. One automated run against generic prompts. Useful for a first look, not for a baseline you will act on.
• Entry audits, roughly RM 2,000 to RM 3,500. Around 20 prompts, two or three engines, a single run, a written fix list.
• Full audits, roughly RM 4,000 to RM 6,000. Thirty to forty prompts, four or more engines, repeat runs, competitor comparison and page-level diagnosis.
• Bundled into a retainer. Often included at no separate charge, with monthly re-sampling built into the reporting.

Treat quotes under RM 1,000 with caution — that is usually a tool subscription with a summary attached. For how audit pricing behaves more generally, see free versus paid SEO audits, and for choosing a provider, how to pick an AI SEO agency in Malaysia.

11. Five Questions to Ask Before You Buy One

Quick Answer: Ask where the prompts come from, how many runs sit behind the numbers, which engines are sampled, whether you keep the raw data, and what the provider refuses to promise. Five honest answers separate a measurement service from a sales exercise.

Take these to the first meeting. Quick to ask, hard to bluff.

1. Where do the prompts come from? The right answer involves your sales team. The wrong answer is a keyword tool export.
2. How many runs are behind each number? One run is an anecdote. Look for at least two or three, spaced out.
3. Which engines, and why those? They should explain why each engine matters for your buyers rather than listing everything.
4. Do I keep the raw prompt-by-prompt data? If you cannot re-run it or hand it to another provider, you are renting the finding.
5. What will you not promise? Anyone guaranteeing a ChatGPT recommendation is guaranteeing something they do not control.

Google’s own guidance on optimising for generative AI features is blunt that there is no special markup or secret setting for AI results. A provider selling a proprietary trick is selling the label. Cross-check the delivery side against who to hire for answer engine optimisation, and the do-it-yourself side against how to show up in ChatGPT answers.

12. Conclusion

Quick Answer: An AI visibility audit is worth buying when it gives you a frozen prompt set, repeat runs, four-level scoring, page-level diagnosis and a date to measure against. Without those, you have paid for an opinion about screenshots.

The vocabulary around AI search will keep changing. The measurement will not: ask real buying questions, write down who gets named, note the date, then repeat. Everything a provider recommends should trace back to that record.

ZenWeb has run search work for Malaysian businesses since 2000. We baseline AI visibility inside the same SEO service rather than selling it separately, because the pages that earn a citation are the pages that earn a ranking — and one team should own both.

13. Frequently Asked Questions

How much does an AI visibility audit cost in Malaysia?

A standalone audit for an SME usually falls between RM 2,000 and RM 6,000, depending on how many prompts, engines and repeat runs are included. Agencies often waive or credit the fee when it leads into an ongoing retainer.

Can I run one myself?

Yes, and it is worth doing once before you pay anyone. Write 20 buying questions in your customers’ wording, run them through ChatGPT, Gemini, Perplexity and Google, and record whether you are absent, mentioned, cited or recommended. The hard part is repeating it consistently.

How often should the audit be repeated?

Monthly for the same frozen prompt set is the practical cadence, with a fuller review each quarter. Answers shift week to week on their own, so anything less frequent makes it impossible to tell real movement from ordinary noise.

Does Search Console show AI Overview data separately?

No. Google states that AI feature appearances are counted inside overall Search Console web search data rather than reported on their own. You can see the traffic effect, but not which answer produced it.

Will an audit guarantee I get recommended by ChatGPT?

No, and any provider promising that is overselling. Nobody controls model output. What an audit does give you is a dated baseline, a clear reason for each gap, and a fix list you can hold the next month’s report against.