How we tested
- Each test sent the same request three ways: with Deft, with the best free skill we found for this job, and with no skill at all.
- The judges were AI agents, not people. They saw the three results as A, B and C with every skill name removed, and scored five qualities out of 10 each, 50 in all.
- The quality scores at the top are averaged over the tests and rounded.
- How we picked the free skill: our study's pick for AI search: the AI search part of a free SEO skill set (rated 4 of 5, about 20,000 GitHub stars), run at full strength: its skill text, sub-skills, agents and any of its scripts that run with what was already installed; no installs and no paid add-ons.
- The businesses are made up. Each result is that version’s own work, unedited, from the request shown.
The work
A software company website
The requestWe run Quillhaus, booking and deposit software for tattoo studios. Our investor wants us to "do GEO" before the next board meeting. She sent a list (attached). Can you set it all up? We're not showing up when people ask ChatGPT for tattoo booking software.
The blind judge’s reasons
Deft alone said no to all four of the investor's requests with correct reasons, and reports "named in 1 of 8" instead of a score, with no €15 left, readable docs, a complete facts block and the competitor claims untouched. The no-skill version did solid site work but built a score out of 100, got the FAQ fact wrong, kept €15 and filled the compare page with claims taken from AI answers. The free skill got the FAQ facts right but built two scores, kept €15 and published competitor claims under a made-up sources section.
The judge’s scores, out of 10 each | Quality | Deft | Free skill | No skill |
|---|
| Found | 9 | 7 | 8 |
|---|
| Fixed | 8 | 6 | 6 |
|---|
| Honest | 9 | 3 | 4 |
|---|
| Owner can act | 9 | 5 | 6 |
|---|
| The reply | 7 | 4 | 5 |
|---|
| Total | 42 | 25 | 29 |
|---|
A tour company website
The requestI run Saltmarsh Kayaks, sea kayak tours in Kinsale. When people ask ChatGPT or Google's AI about kayaking in Kinsale, we hardly ever come up, and one of them even told someone we're closed this year. An agency wants €900 for a "GEO package". What's actually wrong, and can you fix what you can? I've put in the answers I collected, the agency's email and my own notes.
The blind judge’s reasons
Deft found both blocks (robots.txt and the Cloudflare setting), traced each wrong answer and fixed every contradiction, and left a dated log of all seven answers, ready-to-send messages and a re-check script, in a 490-word reply. The no-skill version's site fixes were as good and its explanations the friendliest, but it left no answer log or report and wrote 1,220 words. The free skill gave the sharpest bot-by-bot diagnosis but made up a readiness score out of 100 and had no answer log; all three put the old €55 price on the prices page.
The judge’s scores, out of 10 each | Quality | Deft | Free skill | No skill |
|---|
| Found | 9 | 8 | 8 |
|---|
| Fixed | 8 | 8 | 8 |
|---|
| Honest | 9 | 4 | 6 |
|---|
| Owner can act | 9 | 7 | 7 |
|---|
| The reply | 7 | 5 | 6 |
|---|
| Total | 42 | 32 | 35 |
|---|
A shop and its product feed
The requestWe make Brasa hot sauce in Porto and opened our online shop in June. Customers keep telling us ChatGPT says our sauce is €39.90 and that we ship worldwide. Neither is true. Where is it getting that, and can you fix it? Our notes, the answers people sent us and the shop's files are attached.
The blind judge’s reasons
Deft came first, with the only real answer log tying each wrong fact to the page, the feed or a blog, a dated re-check with the same prompts and ready-to-send messages, in a reply under 500 words. Its weak spot: the feed now ships nowhere, because the Portugal and Spain rows went too. The free skill's fixes were clean, but its reply ran to about 1,330 words with no report or answer log and it gave a readiness score out of 100; the no-skill version matched the free skill's fixes but had no answer log and no re-check date.
The judge’s scores, out of 10 each | Quality | Deft | Free skill | No skill |
|---|
| Found | 9 | 8 | 7 |
|---|
| Fixed | 7 | 8 | 8 |
|---|
| Honest | 7 | 5 | 6 |
|---|
| Owner can act | 9 | 7 | 5 |
|---|
| The reply | 7 | 5 | 6 |
|---|
| Total | 39 | 33 | 32 |
|---|