Teardown 01 · 30 July 2026 · subject: snippettester.com
The first site we tore down was our own.
Before tearing down anyone else's site it seemed only fair to start with the one I am responsible for. This domain used to host a SaaS product — an A/B testing tool for search snippets. The product is discontinued. The landing page was still live, and it is a textbook example of the gap between a page that reads well and a page a machine can use.
Why this page existed, and why it stopped
Snippet Tester was a tool for A/B testing titles and meta descriptions, reading performance back from Google Search Console. It got as far as a working beta with OAuth, scheduling and daily metric sync.
Then it failed its own assessment. Search Console already reports impressions, clicks, CTR and position by page and by date. The tool put a scheduler on top of that and called it a product. It did not do anything Google was not already doing for free, so it was shelved — which was the right call and is worth stating plainly on a site that now sells judgement about search.
The landing page, though, is a useful artefact. It was written carefully. The copy is clear, the structure is logical, a human reads it without friction. And it was close to invisible to an answer engine. Those two things are not in tension — that is the whole problem the category exists to solve.
Finding by finding
| Signal | Score | Verdict |
|---|---|---|
| Structured data | 0 / 18 | Gap |
| Heading structure | 6 / 15 | Gap |
| Extractable formats | 4 / 14 | Gap |
| Evidence density | 2 / 14 | Gap |
| Answer-first opening | 11 / 15 | Partial |
| Metadata hygiene | 4 / 6 | Partial |
| Internal integrity | not scored | Gap |
0 / 18 Structured data
What was there. No JSON-LD and no microdata anywhere in the document. The page had a full FAQ section — six questions with answers, correctly marked up as headings and paragraphs — and none of it was declared as FAQPage.
What I would change. The content already existed in the right shape. Wrapping it in FAQPage plus an Organization block was maybe forty minutes of work for the single highest-value gap on the page.
6 / 15 Heading structure
What was there. Two H1 elements: the brand wordmark in the header and the hero headline. Six question-form headings in the FAQ, which is genuinely good, undercut by a heading hierarchy that starts by contradicting itself.
What I would change. The logo is not a heading. Make the wordmark a link containing an image or a span, leave the single H1 for the hero, and the whole document becomes parseable.
4 / 14 Extractable formats
What was there. Not one table on the entire page. The pricing section compared three tiers across six features each — the most naturally tabular content imaginable — rendered as three stacked lists of bullet points.
What I would change. Pricing comparisons belong in a table with header cells. This is the single most reliably extracted structure on the web and the page had zero of them.
2 / 14 Evidence density
What was there. One statistic on the entire page: "12 sites already testing". No outbound links to anything, no citations, no quoted authority. Every claim was self-asserted.
What I would change. A page about CTR optimisation that cites no CTR research is asking for trust it has not earned. Two or three cited figures would have changed the character of the whole page.
11 / 15 Answer-first opening
What was there. Actually decent. The hero paragraph — "Create variants, schedule runs, and track daily impressions, clicks, CTR and position directly from Google Search Console" — is concrete, declarative and quotable, with no filler.
What I would change. Little to fix. This was the strongest part of the page and it is the part most sites get worst.
4 / 6 Metadata hygiene
What was there. Title 46 characters, description 152 characters — both comfortably in range. No canonical link in the markup. A meta keywords tag, which search engines have ignored since roughly 2009.
What I would change. Add the canonical, delete the keywords tag. The keywords tag does no harm but it tells anyone technical that the page was built from a 2011 checklist.
not scored Internal integrity
What was there. Eleven links pointing at "#" or nothing at all, including both hero call-to-action buttons, the log-in link, and the buttons on two of three pricing tiers. The footer linked to /docs/, /blog/, /guides/ and /support/, none of which had content.
What I would change. This is not an AEO finding, it is the finding. A page whose primary conversion buttons go nowhere has no optimisation problem worth discussing.
The lesson that generalises
The three worst scores on this page — no structured data, no tables, no cited figures — all describe content that already existed in the wrong container. The FAQ was written. The pricing comparison was written. They were just marked up as decoration instead of data.
That is the ordinary case. Most sites do not need more content, they need the content they have to stop hiding. It is also why the free checker measures structure rather than quality: quality is subjective and usually fine, structure is measurable and usually broken.
Method note, and a caveat that matters. The findings above come from the page as served on 30 July 2026, scored against the same eight signals as the free checker. The observations are direct: the two H1s, the absent structured data, the zero tables, the single statistic and the eleven dead links were all read off the page.
The total is a different matter. It was estimated by hand rather than produced by the checker running against the raw HTML, so treat 41/100 as provisional and expect it to move by a few points when re-run. Publishing a hand-estimated number as though a tool produced it is precisely the habit this site exists to argue against, so it is labelled rather than quietly rounded.
Weighting follows the evidence rather than intuition: heading order tracks roughly 2.8x citation odds, structured data is genuinely contested, and llms.txt is scored at zero because 97% of those files are never requested. All of it is argued on the evidence page.