What did two free directory listings send us in 63 days? 29 real sessions from 22 addresses, 2 saw the check form, 1 started a check on launch night, 0 bought

Method, stated before the numbers

Both listings tag their outbound links. Uneed's "Visit" button sends visitors to `/?ref=uneed.best&ref_type=adv` (the `adv` tag is on every tool page on the site, not a paid slot; we checked a control listing on 2026-08-21). Product Hunt sends `/?ref=producthunt`. We queried our PostHog project for every event whose landing URL carried either marker, or whose referrer domain contained `uneed` or `producthunt`, then keyed the result on PostHog's session id and pulled every event in each session, whichever identity emitted it. One session changed identity mid-way when the visitor typed an email into the check form; it counts once. Each session was then run through the viewer exclusions of our standing instrument, `ops/verify/real_events.mjs`, parsed from its source at run time so this cut cannot drift from the tables the rest of the company uses: internal identities, owner IP addresses, adjudicated scanner ranges, fake user-agent patterns and AI-fetcher user agents. The instrument's verdict is reported as-is. Beside it we report one hand-labelled shape with a stated rule: a session is ENGAGED if it emitted an engagement heartbeat, viewed two or more distinct pages, or spanned 30 seconds or more; otherwise it is a BOUNCE. Two sensors are younger than the ledger and we say so in every row: form-in-view (`cta_viewed`) exists since 2026-07-29, visible-seconds heartbeats on the homepage since 2026-08-15. Before those dates the cell reads n/m, not measurable, never no. Script: `content/news/directory_referral_cut.mjs`; artifact with every session's event kinds, pages and clicks: `directory-referral-cut-2026-08-28.json`, 14 of 14 self-checks green. Read at 2026-08-28 13:52 UTC.

The ledger by source

Two things in that table are the finding. First, the two listings are different channels wearing the same label. Product Hunt was a launch-night pulse: 12 of its 15 sessions arrived within 36 hours of launch, 8 of the 15 were single-event sessions of 0 seconds from hosting and cloud ranges (Google Cloud, Linode, IBM, a Kansas City wholesaler) that the instrument does not exclude because none of them is on an adjudicated list, and nothing has arrived from Product Hunt since 2026-07-21. Uneed is a drip: 14 sessions across all 63 days, the newest three in the last eight days, but 8 of the 14 came from one IPv6 /64 in Lagos and only 1 of the 14 came from the US or UK. Second, the one activation the channel ever produced happened on the night the offer was newest, from a visitor who typed rather than tapped: 21 events in 105 seconds, an email entered, a report viewed, no upgrade click. Nothing since has reached a form field.

Month by month

August is the smallest month by count and the only one where we can see the form question at all: 4 sessions, all Uneed, all engaged by the rule, 2 saw the form. The 2026-08-20 Lagos session spent 469 seconds on the site across 2 pages, scrolled to 100%, had the form in view, and touched nothing. The 2026-08-28 Pasig session spent 106 visible seconds on the homepage, scrolled to 100%, had the form in view from second 52, and tapped the footer "Contact" mailto link instead of it. That is the same shape our AI-referral ledger reports: 15 referrals, 5 saw the form, 1 tapped, 0 typed. Two channels with nothing in common upstream are producing the same see-but-do-not-touch outcome at the form, which is why the object on the page, not the source of the visitor, is the variable this company's Sunday deep run is examining.

Every session, in order

All 30 sessions the markers found, including the one the instrument excludes (our own smoke-test identity on 2026-07-02). Span is first event to last event. Visible seconds and scroll are n/m before their sensors existed. "Touch" lists any tap on a check CTA, on the footer Contact link, or on a form field.

What this does and does not show

• The instrument's verdict is a floor, not a human count. Eight Product Hunt sessions are one event and 0 seconds long from cloud ranges; the instrument calls them real because no rule names them, and we report them under that verdict with the shape beside it rather than hand-excluding them. Read the engaged-shape column for the number that resembles people: 10 of 29.
• A network address is not a person. The Lagos /64 that produced 8 Uneed sessions could be one returning reader, a household, or a scanner that keeps a browser profile; its sessions range from 6 to 520 seconds and the 08-27 one scrolled to 100% with 65 visible seconds, which is not a scanner's shape. We count sessions and addresses separately and do not name a person count.
• Form-in-view is measurable on 4 of 29 sessions. The claim "2 saw the form" is 2 of 4, and it says nothing about the 25 earlier sessions, including the 35-minute La Jolla evaluation on 2026-06-27 that read four pages and never started a check.
• Both listings are free. No money was spent on either; `ref_type=adv` is a site-wide tag on Uneed, verified against a control listing.
• The lifetime denominator on this channel is 29 sessions. Rates off it would be noise; every figure above is a count.

Pre-committed read, 2026-09-28

The same script runs again on 2026-09-28 over the same markers, and the artifact is published beside this one. (A) Uneed has sent 14 real sessions in 63 days, about one every 4.5 days; the read stands if it sends at least 3 more real sessions between 2026-08-28 13:52 UTC and 2026-09-28, and "the listing still drips" is falsified below 3. (B) "Directory readers see the form and do not touch it" stands if the directory ledger still shows 0 form-field touches and 0 checks started after this read; one touch or one check falsifies it, and we would rather be wrong. (C) "Product Hunt was a launch pulse" stands at 0 or 1 new Product Hunt session in the 31 days; 2 or more falsifies it. Each is scored on the stated number, nothing else.

If you run a small site and have listed it on a directory, the same read takes one query: find the marker the directory appends to its outbound link, pull every session that carries it, and count sessions, addresses, and form touches separately. If you want the same shape of read on what AI engines say about your business, the free 60-second check is the form this ledger is about.