The most useful question a marketing leader can ask about an AI visibility report has nothing to do with the brand in the report.
It is this. Was this data built to spot a trend, or built to move a budget?
The IAB drew that line formally in August 2026, and it may prove to be the most practical contribution in the entire document. Not because the distinction is novel, but because naming it gives teams permission to ask a question they have been avoiding.
The two tiers
Directional measurement is designed for awareness. Fewer than 50 queries, at least two intent types, monthly or quarterly testing. It tells you roughly where you stand and whether the trend is moving. It is genuinely useful for a leadership briefing or a competitive gut check.
Decision-grade measurement is designed for consequence. Large, diverse query sets covering all four intent categories. Weekly or more frequent testing. Reproducibility with defined variation ranges and stated confidence levels. It is what you need before reallocating spend, restructuring a content program, or telling a board that a competitor is pulling ahead.
The two are not better and worse. They are different instruments for different jobs. A compass and a micrometer are both precise about something. Only one of them is precise enough to machine a part.
The failure mode
Here is what is actually happening in a lot of executive meetings right now.
Directional data is being presented, read, and acted on as though it were decision-grade.
Nobody is lying. The vendor produced a real number from real queries. The analyst put it in a real slide. But somewhere between the tool and the meeting, a figure built from 40 monthly queries acquired the authority of a figure built from 400 weekly ones, and a budget conversation started running on it.
This is not primarily a vendor problem. It is a framing problem, and it is fixable with one line at the top of the report stating what tier the data belongs to.
Six questions to ask your provider
Whether you are evaluating a new platform or auditing the report you already receive, these six questions separate the tiers quickly.
How many queries were run? Volume alone is not quality, but it sets a ceiling on what conclusions the data can support. Forty queries a month cannot tell you much about a market with hundreds of realistic buying questions.
Across how many intent types? Broad category queries, comparison queries, problem-first queries, and constraint-specific queries behave very differently. “Best CRM platforms” and “best CRM for a 200-person B2B SaaS company replacing HubSpot” are separate buying conversations, and a brand can win one while losing the other entirely.
Which systems, and are they blended? ChatGPT, Claude, Gemini, and Perplexity retrieve differently and weight sources differently. A single score averaged across all four without disclosed weights hides more than it reveals.
How often is it run? Model updates, new competitor content, and index refreshes all move results. Monthly sampling will miss changes that matter and misattribute changes that do not.
What is the variation range? This is the question most likely to be met with silence, and it is the most important one. AI answers are probabilistic. A provider reporting a single value with no range is presenting a distribution as a point.
How is a mention defined, and is it separated from a recommendation? Some tools count any brand name appearance. Others count only sourced citations. Since recommendations appear to move behavior two to three times more than mentions, a blended count can shift entirely in composition without changing in total.
Why the variation question matters most
If you only ask one, ask about the range.
Running the same buying question against the same system repeatedly does not produce an answer. It produces a distribution. The IAB guidance says this directly, stating that single-response measurement is not measurement and that providers should report ranges rather than single values.
The width of that distribution carries information a midpoint cannot.
Two brands can share an identical average visibility figure. One appears in nearly every run of the query. The other appears in half of them and is absent from the rest. Those are not equivalent positions. The first has evidence the model holds consistently. The second is sitting near a threshold, present when retrieval happens to surface a supporting source and absent when it does not.
The second brand is not less visible on average. It is less settled. And an average conceals that difference completely.
What to do with directional data
None of this is an argument for discarding directional measurement. Most teams should start there, because most teams are starting from nothing. Roughly 16 percent of brands systematically track AI visibility at all, which means the majority of this market has no baseline of any kind.
Directional data does three things well. It establishes that baseline. It surfaces obvious competitive gaps. And it tells you whether the trend line is moving in a direction worth investigating.
What it should not do is settle an argument about budget. When directional data suggests something significant, the correct response is to escalate to a decision-grade look at that specific question rather than to act on the directional figure.
The one-line fix
Put the tier at the top of the report.
“This is directional data. 40 queries, two intent types, monthly, ChatGPT and Perplexity only.”
That single line changes how every number underneath it gets read, and it protects the analyst as much as the executive. It is the difference between a report that informs a decision and a report that gets blamed for one.
The measurement standard arriving is progress. Labeling your own data honestly is how a team gets the benefit of it immediately, without waiting for the vendors to catch up.
FAQ
What does decision-grade AI visibility measurement mean?
Decision-grade measurement meets a standard high enough to inform budget and strategy decisions. Under the IAB’s August 2026 guidance it requires large, diverse query sets covering all four intent categories, weekly or more frequent testing, and reproducibility with defined variation ranges and stated confidence levels.
What counts as directional AI visibility data?
Directional data uses smaller query sets, typically fewer than 50 queries across at least two intent types, tested monthly or quarterly. It is appropriate for trend spotting, competitive awareness, and internal briefings, but it was not built to support budget allocation.
What questions should I ask an AI visibility vendor?
Ask how many queries were run, across how many intent types, on which systems and whether they were blended, how frequently the testing runs, what the variation range is, and how the tool defines a mention versus a recommendation. A vendor who can answer all six is offering measurement rather than an impression.
Why does the variation range matter so much?
AI answers are probabilistic, so repeated runs of the same query produce a distribution rather than a fixed result. Two brands can share the same average while one appears consistently and the other appears intermittently. The range tells you which situation you are in and the average does not.
Do ChatGPT, Claude, Gemini, and Perplexity need to be measured separately?
Yes. These systems retrieve from different sources, weight recency differently, and require different levels of corroboration before recommending a brand. Blending them into one score without disclosed weights hides where your actual strengths and gaps are.
Is directional AI visibility data useless?
No. It is the right starting point for most organizations, especially since only about 16 percent of brands track AI visibility systematically today. The mistake is not using directional data, it is presenting it as though it were decision-grade when a budget decision is on the table.
Start here: axissuite.ai
Axis Suite by TrendAxis