How to Track AI Search Visibility Without Fake Metrics
The short answer
There is no stable ranking position inside an AI answer, so borrowing rank tracking produces false precision. Measure citation share and consistency against a fixed question set run on a fixed schedule, and separate the inputs you control from the outcomes you do not.
Table of contents
Why rank tracking does not transfer
In classic search there is a results page, positions are numbered, and the same query returns broadly the same list to everyone. That stability is what makes rank tracking work.
AI answers have none of it. Ask the same question twice and you can get different sources. Reword it slightly and the set changes again. Ask from a different account and it changes once more. There is no position three to occupy.
Any tool selling you a ranking position inside ChatGPT is manufacturing a number that does not exist. The variance is a property of the system, not a measurement error to be smoothed away.
Start with a fixed question set
Everything downstream depends on this. Build thirty to fifty questions a real buyer would ask, write them down, and then do not change them. A moving question set produces movement that means nothing.
| Category | Share | Example shape |
|---|---|---|
| Problem framing | ~30% | Why is our organic traffic falling |
| Solution research | ~25% | How do businesses improve AI search visibility |
| Provider selection | ~20% | Who are the best AI SEO consultants in Malaysia |
| Comparison | ~15% | Agency versus in-house for search work |
| Brand and objection | ~10% | Is this consultancy any good, what does it cost |
Include several rewordings of the same underlying question. Consistency across phrasings is a stronger signal than one lucky hit, and you can only see it if you deliberately test for it. The mix should reflect how buyers actually phrase things, which is the practical difference explored in GEO vs SEO.
The metrics that hold up
| Metric | How to calculate | Why it holds up |
|---|---|---|
| Citation share | Questions where you are cited, divided by total questions run | Direct measure of presence. Comparable across months |
| Citation consistency | Of the questions where you appear, how many rewordings you survive | Separates real trust from a one-off retrieval |
| Competitive share of voice | Your citations against each named competitor's, same question set | Context. Everyone rising together means the category changed, not you |
| Trusted-domain overlap | How many recurring cited domains also mention you | Leading indicator. Moves before citation share does |
| Editorial mention velocity | New third-party mentions per month | An input you control. Proves the programme is shipping |
| Branded search volume | Search Console, brand queries, month over month | Catches influence that produced no click |
The last two matter more than they look. AI visibility often shows up first as people searching your brand name directly, having encountered you in an answer they never clicked. If you only count sessions from AI referrers, you will conclude nothing is happening while it is happening.
The monthly routine
This takes about ninety minutes a month done manually. That is cheap for the only measurement in this field that survives scrutiny.
- Same day each month. Pick one and hold it. Comparability comes from consistency in the method, not sophistication in the tooling.
- Clean sessions. Logged out, no chat history, location set to your actual market. Personalisation will otherwise flatter you with results nobody else sees.
- Run the full set across your priority platforms. For most Malaysian businesses that means ChatGPT, Perplexity and Google AI Overviews. There is a fuller comparison in which AI search engines actually send business traffic.
- Record cited domains, not impressions. One row per question per platform, listing every domain cited. Normalise to root domains.
- Calculate the six metrics. A spreadsheet is sufficient. Nobody needs a dashboard for thirty rows.
- Note what changed. New competitor appearing, a publication suddenly dominating, your own pages appearing for the first time. The qualitative notes are usually where the insight is.
Tools exist that automate parts of this. They are useful for volume and they still need a fixed question set to be worth anything. The method is the product. The tool is convenience. This measurement model is the reporting layer inside my AI SEO and GEO service, not a separate deliverable.
Separate what you control from what you do not
This is the discipline that keeps a reporting relationship honest, and it is the one most often skipped.
| Controlled inputs | Uncontrolled outcomes |
|---|---|
| Editorial mentions earned | Citation share |
| Pages restructured to answer directly | Citation consistency |
| Schema and technical work shipped | Competitive share of voice |
| Evidence assets published | Whether any single query cites you today |
Nobody controls whether a given question cites you on a given day. What is controllable is the quality and volume of the work that makes citation more probable. Reporting both, clearly labelled, means a flat month does not look like a failure and a good month does not get overclaimed.
It also protects against the opposite failure, where inputs are all anyone reports and outcomes never get examined. Both columns, every month.
Frequently asked questions
How do you measure AI search visibility?
Run a fixed set of thirty to fifty buyer questions on a fixed monthly schedule and record which domains get cited. From that, calculate citation share, consistency across rewordings, and competitive share of voice. Consistency of method matters more than any specific tool.
Can you track rankings in ChatGPT?
No, because there is no stable ranking to track. The same question can return different sources on repeat asks, and rewording changes the set again. Tools reporting a position inside ChatGPT are producing false precision. Measure how often you are cited across many attempts instead.
How often should I check AI visibility?
Monthly is the right cadence for most businesses. Weekly checks mostly capture noise from natural variance rather than genuine movement. Run the same questions on the same day each month, logged out and with location set to your actual market.
What is a good citation share?
There is no universal benchmark, because it depends entirely on category competitiveness. The useful comparison is against your own baseline and against named competitors on the same question set. Moving from zero to any consistent presence is the meaningful first milestone.
Why is branded search volume an AI visibility metric?
Because many AI answers influence a buyer without generating a click. Someone reads a summary naming you, then searches your brand directly later. If you only count referral sessions from AI platforms, that influence is invisible, and you will underestimate what is working.
Do I need a paid tool to track AI visibility?
Not to start. A spreadsheet and ninety minutes a month covers a thirty-question set across three platforms. Paid tools help with larger question sets and repeated sampling, but they still depend on you defining a stable question set, which is the part that actually determines usefulness.
Written by Robin Ooi
Robin is a Malaysian AI SEO and reputation specialist with more than fifteen years in search. He is the Amazon bestselling author of Your SEO Sucks! and works with MNC and listed company clients across Kuala Lumpur, Penang, Johor and Singapore. Read more about Robin Ooi.
