There are two distinct ways to be invisible in AI search, and they have opposite fixes. Either the engine never names your brand in answers to your customers' questions (a brand-mention gap), or it covers your topic but never cites your pages as the source (a source gap). Measuring only one hides half the problem. This two-lens audit — part 7 of the measuring series — separates the demand side from the supply side so your fixes land where the gap actually is.
What are the two lenses?
Each lens answers a different question about the same set of AI answers:
| Lens | The question it answers | What a low score means |
|---|---|---|
| Brand-mention gap | Across my customers' prompts, how often am I named or recommended? | Demand-side: weak presence, reputation, or entity strength |
| Source gap | When the answer covers my topic, how often are my pages cited as the source? | Supply-side: weak extractability, reachability, or freshness |
The first lens is about share of voice and total mentions — am I in the conversation at all? The second is about citation coverage — when the conversation happens, is my content the supply that feeds it? They move independently, which is the whole point.
Why measure both separately?
Because the same overall "we're invisible" symptom has two unrelated causes, and treating the wrong one wastes a quarter. The four combinations:
- Named and cited — healthy; defend it with freshness.
- Cited but not named — your page feeds the answer, but a competitor is recommended. This is the cited-isn't-recommended pattern: an entity or corroboration problem, not a content one.
- Named but not cited — the model "knows" you from training, but live retrieval pulls others' pages. A content extractability and freshness problem.
- Neither — the hardest case; start with reachability and entity basics.
Knowing your overall AI visibility is "low" tells you nothing actionable. Knowing whether the gap is demand (you're not named) or supply (you're not cited) tells you exactly which team owns the fix.
How do you perform the audit?
Snapshot the same prompt set across engines on a schedule and score each lens. A workable method:
- Build the prompt set. 30–50 real customer questions across the fan-out branches — brand, category "best X," comparison, and how-to intents. Choosing the right prompts is most of the value here.
- Snapshot across engines. Capture answers for each prompt on ChatGPT, Google AI Mode, Perplexity, and others — repeated captures, not one, since answers are non-deterministic.
- Score lens 1 (mentions). For each prompt, was your brand named? Recommended? Compute a mention rate and a recommendation rate per engine.
- Score lens 2 (sources). For prompts about your topic, what share of cited sources are your pages versus third parties versus competitors?
- Map each gap to its fix. Mention gaps → entity clarity, corroboration, earned coverage. Source gaps → extractable answer-first pages, reachability, and a freshness cadence.
What do you do with the result?
Rank fixes by which lens is weakest and which prompts are highest-intent. A brand strong on sources but weak on mentions should invest in entity strength and earned placement, not more content. A brand named everywhere but rarely cited should fix extractability and freshness on existing pages. Re-snapshot quarterly and watch both scores trend — a source score that slips is the early warning that pages are decaying out of answers. Tracking both lenses continuously, across every engine, with fixes prioritised by impact, is exactly what Buffy Intel measures.
