The Complete Measured AI Visibility Guide for Enterprise Buyers

Why Enterprise Buyers Get This Wrong

Enterprise procurement teams are good at evaluating SaaS vendors. They are unfamiliar territory for AEO vendors because the category is new and the metrics are unfamiliar.

The result is predictable. Vendors get hired on the strength of a slide deck. The first quarterly business review goes well. The second one struggles when finance asks how the numbers were measured. The third one becomes an exit because the vendor cannot defend the methodology under audit.

This guide is the vendor evaluation framework that prevents that pattern. It is what the strongest enterprise procurement teams already use, written for the ones still building the muscle.

The Four Buckets That Matter

Every measured AI visibility vendor should be scored on the same four buckets. The buckets are weighted differently depending on the buyer's industry, but the buckets themselves do not change.

A vendor that scores high on all four buckets is rare. A vendor that scores high on two of four is the modal vendor. A vendor that scores high on zero is the one most enterprise buyers fund first because the deck is best.

Methodology in Detail

The methodology question is the one most vendors fail.

Ask any prospective vendor four sub-questions. What prompt set is used to generate the conversations that get measured. How is the prompt set selected and how often is it refreshed. How are mentions detected and classified. How is positioning within a recommendation list scored.

A defensible answer to each is a single page of documentation, not a sentence. Vendors who cannot produce that documentation are not measuring. They are estimating.

Coverage in Detail

Coverage means the platforms and the buying topics.

Platform coverage should include at least ChatGPT, Claude, Gemini, and DeepSeek for almost every B2B enterprise category. Some industries also require Perplexity. Single-platform measurement is not AI visibility measurement. It is single-platform measurement marketed under a broader name.

Topic coverage means the prompts the measurement covers actually map to the buying journeys the enterprise cares about. A vendor measuring 200 generic prompts may report a higher score than a vendor measuring 30 enterprise-specific prompts. The 30 is more useful to procurement.

Attribution in Detail

Attribution is what separates measurement from observation.

The right test: ask the vendor to describe a citation share change in the last six months and what specifically caused it. A strong vendor names the intervention, the timing, and the counterfactual. A weak vendor names the trajectory and credits the program in general.

Without attribution, score changes are not actionable. The CFO eventually notices and cuts the line item.

Audit Defensibility in Detail

Audit defensibility is the requirement most overlooked by buyers and most easily failed by vendors.

The questions that matter: is the methodology versioned, when did it last change, are old reports still reproducible under the older methodology, can the vendor produce the raw conversation data on request, and is the prompt set documented and stable.

Procurement teams that ask these questions before signing avoid the painful 18-month conversation that ends with the vendor unable to reproduce a baseline because the methodology silently changed.

Building the Scorecard

A working enterprise scorecard for measured AI visibility vendors looks like this. Methodology and audit defensibility weight 60 percent combined. Coverage weights 25 percent. Attribution weights 15 percent. The bar to pass is 75 percent overall and 60 percent in every bucket.

Vendors below 60 percent in any bucket are disqualified regardless of overall score. The asymmetry matters. A vendor strong in methodology but weak in coverage delivers half a measurement program. A vendor strong in coverage but weak in attribution delivers an unactionable dashboard.

What Procurement Often Skips

Three checks that procurement teams often skip and then regret:

Reference calls scoped to data integrity, not satisfaction. Most reference calls ask "are you happy with the vendor." The right question is "have you ever been unable to reproduce a metric the vendor reported, and what was the resolution."

Sample raw output review. Ask for an unredacted week of raw AI conversation captures. If the vendor cannot produce them, they are not capturing them. They are inferring from aggregated stats.

Methodology change history. Ask for the changelog of the scoring methodology in the last 12 months. A short, versioned changelog is healthy. No changelog at all is a red flag.

How OnlyAEO Scores Itself

OnlyAEO uses Gumshoe for primary measurement. Gumshoe documents its methodology in a single public document, versions changes, retains raw conversations for re-scoring, and covers ChatGPT, Claude, Gemini, DeepSeek, and Perplexity. We use the framework above on our own measurement layer because procurement teams already use it on us. The framework holds up because it is the same one we recommend to buyers.