AI visibility gap analysis: how to find what the engines leave out
The short answer
A gap analysis measures one distance: between where you should be cited and where you actually are. You write down the questions on which being absent costs you something, measure what the engines return on exactly those questions today, and the difference is the work list.
It is not the same exercise as monitoring. Monitoring answers is this improving? A gap analysis answers what do we fix first? You run the analysis once, before any work starts, and it produces the baseline that monitoring then measures against.
Four gaps are worth separating, because each points at different work.
The four gaps
| Gap | The question it answers | What it usually means |
|---|---|---|
| Presence gap | On which priority questions are you absent from the answer entirely? | The engines have no reason, or no usable source, to include you |
| Share gap | Where you do appear, how much of the answer space do competitors hold? | You are in the category but not among its stronger evidence |
| Accuracy gap | Where what is said about you is wrong, stale or conflated | Bad or outdated sources are being read as authoritative |
| Source gap | Which domains the answers are actually built from, and how few are yours | The work to be done, stated as a list of sources |
The source gap is the one that makes the other three actionable. Presence, share and accuracy describe symptoms; the sources behind each answer tell you where the engine’s view of you actually comes from, and therefore what to change. An analysis that reports the first three without the fourth has told you that you have a problem without telling you where it lives.
Running one
1. Write the prompt set. The commercial questions a buyer puts to an assistant before shortlisting in your category. Not brand searches: if someone already types your name, their query cannot tell you whether the engine would have surfaced you unprompted. A few dozen questions is enough to be representative.
2. Define what good would look like. For each prompt, who would a fair, well-sourced answer cite? This step is uncomfortable and it is the one most often skipped, because it forces an honest view of whether you genuinely belong in a given answer. A gap only exists where you had a reasonable claim to be there. Recording an absence as a gap on a question you have no business answering inflates the report and wastes the work.
3. Measure, repeatedly. Run the set across the engines your buyers actually use, and run it more than once, because the same prompt can answer differently an hour later. Capture the full answer, not a verdict: the mention, the citation, what was said about you, and every source domain behind it.
4. Subtract. Set the readings against step 2, and sort what is left into the four gaps above.
For the underlying observation method, see how to measure AI search visibility; for the per-engine specifics of Google’s surfaces, which behave differently enough to need separate readings, see how to measure AI visibility in Gemini.
Reading the output
The value of separating the gaps is that each points somewhere different.
An accuracy gap is the one to move on first, whatever its size. Absence is a missed opportunity, but a confident wrong answer about your ownership, your pricing or your positioning does damage every time it is served, and for a regulated, listed or contested organisation it is an exposure rather than a marketing problem. Accuracy gaps are also often the cheapest to close, because they usually trace to a small number of identifiable stale sources.
A presence gap on a high-intent question is worth more than several on low-intent ones. Weight the list by how close each question sits to a decision rather than by how many prompts are affected.
A share gap is usually a symptom of the source gap rather than an independent problem, so resist treating it as its own project. As the source work lands, share tends to follow.
A source gap is the work list. It names the pages of yours that are not being selected, and the third-party sources being selected instead.
What a gap analysis will not tell you
It will not tell you how long the gaps take to close. Source-layer fixes have to be published, then found, then chosen, and every step of that runs on the engine’s schedule rather than yours. Anyone offering a guaranteed timeline for AI visibility is selling certainty they do not have.
It will not, on its own, change anything. The analysis is a diagnosis; the improvement comes from the source work it points to, and the proof comes from re-measuring the same prompts afterwards. This is why the prompt set has to be written down and frozen at this stage rather than reconstructed later: it is the only thing that makes the second reading comparable with the first.
And it will not survive being run once and filed. Answers shift as models retrain and competitors publish, so a gap analysis ages. It is a starting point for a standing programme, which is the argument for managed monitoring rather than a repeated series of one-off reports.
Where Morris McLane fits
Our AI visibility audit is this analysis, run for you: your prompt set built from the questions your buyers actually ask, read across the engines that matter to them, checked by hand rather than scraped, and returned as the four gaps with the two or three fixes that would move them. It is a fixed fee, quoted before we start.
Where it justifies a standing cadence, the audit becomes the baseline for AI visibility monitoring within our AI search visibility work, for organisations directly and white-label for communications and government-relations firms reporting to their clients.
Frequently asked questions
What is an AI visibility gap analysis?
A one-off diagnostic that measures the distance between where an organisation should appear in AI answers and where it actually appears. You define the questions on which being absent costs you something, measure what the engines currently return on exactly those questions, and the difference is the gap. Unlike ongoing monitoring, which answers whether things are improving, a gap analysis answers what to do first. It produces a work list, not a scoreboard.
How do you run an AI visibility analysis?
In four steps. Write the prompt set: the commercial questions your buyers put to an assistant before they shortlist, not brand searches. Define what good would look like on each prompt, meaning who you would expect a fair answer to cite. Run the set across the engines your buyers use, repeatedly rather than once, recording citations, mentions, accuracy and the source domains behind each answer. Then subtract: where you are absent, where you are present but outweighed, where you are described wrongly, and whose sources are being used instead of yours.
What is the difference between a gap analysis and ongoing AI visibility monitoring?
Purpose and cadence. A gap analysis is a one-off read that establishes where you stand and what to fix, and it is most useful before any work starts. Monitoring is a standing measurement on the same frozen prompt set that tells you month over month whether the fixes are landing. The gap analysis produces the baseline that monitoring then measures against, so they are sequential rather than alternatives: diagnose once, measure continuously.
How can I tell if an AI visibility solution is actually improving mentions?
Insist on a before-and-after on an unchanged prompt set. If the questions were frozen at baseline and the same questions are run on the same engines each cycle, then movement in citation rate, share of citations and accuracy is real movement. If the prompt set changed between readings, or was never written down, or the provider reports a composite score you cannot decompose into prompts and engines, you cannot distinguish improvement from a change of measurement. Ask which prompts moved, on which engine, and which sources changed behind them.
What do you need before running a gap analysis?
Mostly clarity about your own commercial reality rather than any tooling. You need the questions that precede a purchase or a shortlist in your category, an honest list of the competitors a fair answer might reasonably cite alongside you, and the pages you consider authoritative on each claim. Without the first, you measure the wrong prompts; without the second, you cannot read the share gap; without the third, you cannot tell whether an engine ignored your evidence or you never published it.
Which gap should you fix first?
Accuracy gaps, then presence gaps on the highest-intent prompts. A confidently wrong answer about your ownership, pricing or positioning does active damage every time it is served, so it outranks absence. After that, presence on the questions closest to a buying decision is worth more than presence across a wider set of low-intent questions. Share gaps are usually a consequence of the source gap rather than a separate problem, so they tend to close as the underlying source work lands.
How long does it take to close an AI visibility gap?
Longer than a content change and shorter than a rebrand, and it is genuinely uncertain because you do not control the retraining or re-crawling schedule of any engine. Source-layer fixes have to be published, then found, then chosen, and each step runs on the engine's timetable rather than yours. This is exactly why the closing measurement matters: the honest answer is that you re-measure on the same prompts until the reading moves, and treat any provider offering a guaranteed timeline with suspicion.
