The ad industry's trade body has written down what counts as a real measurement of AI visibility, and a 25-question board sits one rung below its bottom rung.

The Interactive Advertising Bureau released a 36 page document defining a shared vocabulary for brand visibility in AI answers. It organises metrics into four layers it calls the 4 P's: Presence (mention rate, citation rate, share of voice), Prominence (placement and ranking), Portrayal (sentiment, framing, hallucination and factual inaccuracy rates), and Persuasion (recommendation strength and post-citation click-through). IAB's stated reason for writing it is that more than 20 vendors sell AI visibility tools using different query sets, different platform coverage and different scoring rubrics, and they disagree with each other about the same brand.

The load-bearing half is a two-tier quality grading. Directional measurement identifies patterns and signals trends and is explicitly not sufficient for budget allocation. Decision-grade measurement has to clear nine criteria covering query volume, sample size, prompt type coverage, testing cadence, reproducibility and platform coverage. Two of those criteria carry numbers. AdExchanger and PPC Land independently report a hard floor at 50: fewer than 50 queries is graded "exploratory", below even directional, and the guidance's own wording is that fewer than 50 prompts "cannot meaningfully characterize a category". Directional needs at least two of four prompt intent types (informational, comparison, recommendation, transactional) and decision-grade needs all four. On cadence, monthly or quarterly is enough for trend monitoring while budget allocation needs weekly or more frequent testing.

Two caveats have to travel with this. IAB itself says this is not a standard, because the space is not stable enough for one, and that it deliberately avoided calling it a framework; it is guidance for a transitional period. And the numeric thresholds above come from two trade press readings of the PDF rather than from a quoted page of the document, so treat the exact figures as well-corroborated secondary reporting rather than a verified primary quote.

Somantra, an AEO platform, analysed 2,437,107 citation records across 28,725 unique domains on ChatGPT and Google, restricted to the Australian insurance category, across seven observed months between November 2025 and July 2026. It found 57.2% of domains appeared in exactly one sampled month and 2.7% held a citation in all seven. It also reports that content labelled a "complete guide" was three and a half times more common among domains that vanished after one citation, while comparison-format pages, FAQ pages and how-to pages correlated with persistence.

The caveats are serious enough that the number cannot be quoted bare. The seven months are non-consecutive (April and June are missing), so "never cited again" actually means "not re-observed inside a gapped window". The survival table sums to 10,446 domains rather than 28,725, so which population the percentage describes is unclear. Monthly ChatGPT citation volume in the same dataset swung roughly 29x, which alone would produce a lot of apparent one-month appearances. The scope is one country and one industry, and the publisher sells the tracking software. Syndicated versions of the release strip the Australian insurance qualifier entirely.