78 people started our free AI visibility tool over 90 days. 26 finished it. 10 became leads.
That is 12.8% end to end, and 38% of everyone who saw a score. We would take those numbers on any lead magnet.
In the same 90 days, the tool’s own page received 26 sessions.
The tool is not the problem. Nobody is looking at it.
The funnel
Ninety days, from GA4.
| Step | Events | Users |
|---|---|---|
| Audit started | 78 | 49 |
| Score reached | 26 | 22 |
| Became a lead | 10 | 10 |
- Start to finish: 33%. Two thirds begin and abandon before a score.
- Finish to lead: 38%. Of the people who see a result, nearly four in ten give us their details.
- End to end: 12.8%.
The second number is the one that says the offer is fine. Someone who has just been told their site scores 66 out of 100 wants to know what to do about it, and a reasonable share of them will pay the price of an email address to find out.
The first number is where the work is. Two thirds drop out mid-tool.
The part that actually went wrong
For most of its life the tool gave wrong answers to the best-optimised sites.
Two of the checks used regular expressions that assumed HTML attributes would be wrapped in quotes,
like type="application/ld+json". Most production sites serve minified HTML, and minifiers strip
quotes that are not strictly required. So on any minified site the structured-data check and the
meta-description check both failed, regardless of whether the markup was there.
The sites most likely to be minified are the sites most likely to be well built. The tool was hardest on exactly the people whose markup was correct.
We found it by running the tool against our own site. It scored 66, “Emerging”. The real score was 79, “Authority”, with the structure component going from 10 to 19 once the checks worked.
Nobody reported it. They would not: if a free tool tells you your schema is missing, you assume the tool is right and you are wrong.
Run your own tool against your own site, and against a site you know is well built. We should have been doing that monthly from the start.
What we are doing about the traffic
The diagnosis is a tool that only converts people who already found us, which makes it a conversion asset rather than an acquisition one.
The fix is to put the tool where the problem is already being discussed, rather than promote the tool harder. We hold 42 queries at page one earning zero clicks where someone is describing exactly the thing the tool measures: why their brand is missing from AI answers, why a competitor appears and they do not, how to fix wrong information about their company. The biggest single one carries 420 impressions at position 1.4.
Those pages exist. The tool is not on them. That is a linking job, not a marketing campaign.
If you run a free tool
Separate the two failure modes before you fix anything. A low completion rate is a tool problem: too long, too much friction, the result not worth the effort. A healthy completion rate with a low lead count is a traffic problem. They look identical on a dashboard that only reports leads, and they need opposite responses.
Gate the result, not the tool. Let people run it and see something, then ask for the email to get the detail. Our gate is seen by very few people relative to starts, which tells us the drop-out is happening inside the tool rather than at the ask.
Put it where the problem is discussed, not on a tools page nobody visits. A tools page is a destination for people who already trust you.
Test it against your own site every month. Ours was wrong for months and the only reason we caught it was deciding to score ourselves.
Related
Part of a series where we publish our own numbers including the unflattering ones: 6,482 page-one rankings that produced 218 clicks, the estate audit that measured one search engine, organic at 9% of our traffic and 86% of our leads, and how long AI citation actually takes.
The tool itself, now with working checks, is the free AI Visibility Score.
Common questions
Do free tools work as lead magnets?
Ours does: 78 people started it over 90 days, 26 finished, and 10 became leads. That’s 12.8% end to end and 38% of finishers converting.
The problem was never the conversion rate. The tool’s own page took 26 sessions in the same period, so the mechanism works and almost nobody reaches it.
What conversion rate should a free tool get?
Ours runs 33% from start to completion and 38% from completion to lead, so 12.8% overall. We’d treat anything above 10% end to end as working.
If your rate is far below that, the problem is the tool or the gate. If your rate is fine and your lead count is low, the problem is traffic. Those need completely different fixes.
Should I gate a free tool behind an email form?
Gate the result, not the tool. Let people run it and see something useful, then ask for an email to get the detail.
Our gate is seen by very few people relative to starts, which tells us most drop out before the result rather than at the form, so the friction is in the tool rather than the ask.
Why did your free tool give wrong answers?
Two regular expressions assumed HTML attributes would be wrapped in quotes. Most production sites run minified HTML, where the minifier strips them, so the checks for structured data and meta descriptions failed on exactly the sites that were best optimised.
We found it by running the tool against our own site, which scored 66 when it should have scored 79.
How do you drive traffic to a free tool?
Links from the pages where the relevant problem is already being discussed, and the tool having a reason to exist on a search result of its own. Ours had neither for most of its life.
A tool that only converts people who already found your site is a conversion asset rather than an acquisition one.