Where does a free AI visibility check lose people? 34 people saw the form in 30 days, 5 clicked into it, 2 typed, both ran a check within 2 minutes of the first click, and 1 of 6 reports ever led to a paid click

The question and the instrument

A practitioner running a free tool on a site wants one number per step: how many people saw the form, how many touched it, how many typed, how many finished, how many opened the result, how many paid. Our site has carried sensors for each step since the dates in the table below, and every count here is of real external persons: our own identities, our own network egress, our probe loads and two named datacenter shapes are excluded before anything is counted. The last read on this funnel was a lifetime count over 74 days with no keystroke sensor. This one is the trailing 30 days with a keystroke column, a step-1 column and the paid-plan card, and it lands the day after two strangers ran a check on the same afternoon for the first time since 2026-08-28.

Trailing 30 days, by surface

Persons, not events. A person who had a form in view on two surfaces counts once in the pooled line (2 did). "Touched" is a click into any field of the form on that surface. The step columns start at the click: typed, step 1 and check are counted among the people who clicked in.

The /news line's one check is the 2026-08-28 reader who clicked into a form under a post and ran the check a minute later; the step-1 sensor was born four days after that, so the step funnel counts 2 checks where the see-to-touch line counts 3. Nobody ran a check without having had a form in view, and nobody typed or completed step 1 without a click into a field first (the instrument prints all three cross-checks at 0). Held out by the datacenter class: 2 homepage sessions, 4 across all landings.

Every click into a form since the sensor was born

5 people have clicked into a check form since 2026-08-04; every first click landed on the first field, the business name. 3 of the 5 ran a check. Of the 2 clicks after the keystroke sensor went live on 2026-09-10, 2 typed and 1 did not: the 2026-09-09 reader who clicked into all three fields in three seconds, four of the clicks flagged dead by the analytics vendor, and never typed a character.

The two checks of 2026-09-14, minute by minute

Both readers used the homepage form. One arrived on a desktop from a ChatGPT answer that linked our guide on how to check whether AI recommends a business, read the about page, then the homepage; the other arrived on a phone with no referrer. The minutes below are from the event timestamps.

From the first click to a report on screen took 1 minute and 3 minutes. Before the first click, reader 1 spent 62 minutes on the site with the form in view twice; reader 2 spent 3. Both opened the report a second time within ten minutes (our analyst's read of the same tape traces a load on each to the report email), and neither clicked the paid plan.

After the report: every real check since June

6 of 6 reports were opened, 4 of 6 more than once, 1 of 6 led to a click on the paid audit and 1 to a checkout session (2026-07-13; payment is read from the payment processor, not from this tape). The fix-plan card that every stranger's report has carried since 2026-09-11 has been in view on 4 report loads by 2 people and clicked 0 times. That read is pre-registered by our own daily memo on its own denominator (at least 6 people, or 2026-10-09), so this post reports the count and does not score it.

Where the loss is

• Before the first click: 29 of the 34 people who had a form in view in 30 days never clicked into it (85.3%). On /news posts it is 11 of 13; the ChatGPT-referred readers of one post showed the same shape, 7 sessions, 1 click, 0 checks.
• Inside the form: 2 of 2 people who typed completed step 1 and ran a check, in 1 minute or less from the click to step 1. The one click that produced no keystrokes is the 2026-09-09 dead-click session, already published.
• After the report: 5 of 6 people who opened their report never clicked the paid plan, and 4 of those 6 came back to the report at least once, so the report is read and the offer under it is not taken.

What this read does not show

5 clicks and 2 keystroke sessions in 30 days are a handful; the percentages above are counts written as percentages, not rates. The keystroke sensor is five days old, so the "typed" column exists for 3 of the 5 lifetime clicks. The 30-day window holds 3 of the 6 lifetime checks. The report-page engagement sensor (visible seconds and scroll depth on the report) shipped this morning and has no rows on any real report view yet, so whether a reader who does not click the paid plan ever scrolled to it is not readable on this tape. Business names, emails, cities and IPs are not in the artifact.

The next read, pre-registered

1. 1On 2026-09-29, among real external persons whose first click into any check form lands after 2026-09-15 16:00Z: at least 3 of every 4 who type run a check. Fewer than that and the form is a wall after all, and the post says so.
2. 2On the same date, the about page's form (live since 2026-09-15 12:37Z, 0 rows so far) and the two /for pages' hero forms (live since 2026-09-13, 0 rows so far) print their own see-to-touch lines beside the homepage and /news lines above.

Either outside its statement is published as a falsification with the same script and a dated artifact.

Artifact

`content/news/check-funnel-cut-2026-09-15.json` holds the 30-day lines by surface with the instrument's basis strings, the lifetime click, keystroke and step rows by day and surface, every real check's report loads and clicks, the two 2026-09-14 timelines in minutes, three raw query counts beside the instrument's, and 19 self-checks. The script is `check_funnel_cut.mjs` beside it.