What 25,220 AI answers, 1,943 website scans, and 1,253 buying signals reveal

The three corpora

Separate datasets, one operating question

Where is a commercially useful signal hiding, and what evidence would make it actionable? These snapshots answer that question from three different points in the customer journey.

Finding 01 · AI visibility

AI recommendations remain fragmented

MentionedOn collected 25,220 answers and organized them into 1,261 local leaderboards spanning 21 trades and 40 U.S. cities. The corpus contained 8,299 distinct businesses. On average, the leading business in a leaderboard appeared in just 24% of its sampled answers, while 91% of businesses appeared in only one city.

The practical lesson is not that one company “wins AI.” Recommendations vary by category, geography, model, and prompt. A business needs market-level measurement, source evidence, and repeated checks—not a single vanity query.

Source: MentionedOn State of AI Search. The sample measures the prompts, markets, and collection period described in that report; it is not a census of all AI answers.

Finding 02 · Website operations

Pre-consent tracking is common operational behavior

Across 1,943 completed RegSentry scans, 1,106 sites—57% of the sample—loaded at least one tracking technology before the scanner observed a consent choice. Those affected sites produced 2,511 findings, an average of 2.3 each, and 351 sites produced three or more. Meta Pixel was observed on 458 sites, roughly 24% of the full sample.

This is an implementation signal: what a browser observed loading, when it loaded, and whether a consent interaction had occurred. It can help an owner, developer, or legal adviser find configuration work. It does not determine that a site violated a law.

Source: RegSentry State of Website Tracking. Results are technical observations from the scan conditions and definitions published in the source report.

Finding 03 · Sales timing

Buying intent becomes more useful when signals stack

Intakra recorded 1,253 public market signals across 806 watched companies. Of the 459 companies with at least one signal, 129 showed two or more signal types—28.1%. Hiring, funding, and product-launch activity accounted for 970 signals, or 77.4% of the corpus.

A public event is a reason to investigate, not proof that a company will buy. Multiple relevant signal types can give a seller a more specific, source-cited reason to act now, but frequency alone does not establish purchase intent or conversion probability.

Source: Intakra State of Buying Signals. The report documents the categories, sources, and snapshot limits behind these totals.

From signal to action

Use the right evidence at the right operating moment

Read the limits

What these findings do not prove

• The three corpora do not share a denominator and cannot be combined into a single market statistic.
• MentionedOn measures its disclosed prompts, models, trades, cities, and collection windows—not every possible AI recommendation.
• RegSentry reports technical observations, not legal conclusions, and a scan can change as a site or consent flow changes.
• Intakra records public events that may justify research or outreach; an event does not prove a company intends to purchase.
• All figures are a dated snapshot. Follow the source reports for current totals and full methodologies.

Soxoa operating model

Turn an invisible process into a measurable system