Case study: how our own site went from 87 to 100 in 2 days

Short answer: this case study covers 2 days of work on searchbind.com, a brand-new domain. Our own scanner scored the site 87, then 91, then 93, and after one more round of fixes 100 out of 100: all 59 applicable checks pass. Part of the gain came from fixing our site and part from correcting the checklist itself. In a one-time AI visibility check of 10 buyer questions, we appeared in 1 and were linked in none.

The starting point

SearchBind launched in October 2026 on a brand-new domain. No backlinks, no reviews, no history. That makes it a fair test: most of our customers are small sites with little authority, and this is what such a site looks like on day one.

We treat our own site as customer number zero. We run the same scan any visitor can run on the free audit page, and we publish the result here, including the checks we fail.

The numbers

Four runs over two days. The scanner changed between the runs, so the scores are not a clean before and after. The table shows exactly what each run measured.

"Not evaluated" means the check did not apply or could not be measured on that run. On the last run both were not applicable: B9 (LocalBusiness schema; we are not a local business) and E9 (hreflang; the site has one language). D2 is one of the 10 high-impact checks that are explained in paid reports. The other codes are on the public 61-point checklist.

What changed between the runs

87 to 91: we corrected the checklist

The first run used the first version of the scanner, with 43 checks. Running it on our own site showed that parts of the checklist needed correcting. We corrected them and added new checks, which took the scanner to 59 checks the same day. Because the checklist changed between the two runs, the move from 87 to 91 says more about the scanner than about the site. We say this plainly because a jump in a score you control yourself is easy to oversell.

The 59-check run failed five checks. Four of them are public: no profile links in our schema (B2), no person as author on the compare page (C9), no use-case pages (D4) and no case study with numbers (D6).

91 to 93: profile links and two email checks

We fixed B2 by adding sameAs links to the profiles we actually run. The same three links sit in the Organization schema on the home page and in the founder's Person schema on the about page:

"sameAs": [
  "https://x.com/okarabuilds",
  "https://www.threads.com/@okarabuilds",
  "https://www.reddit.com/user/Puzzleheaded-Arm5742/"
]

We also added two checks to the scanner, for SPF (E19) and DMARC (E20). Forged email from a domain hurts trust in that domain, and both records take minutes to set up. Our domain passes both. These are the records as published today:

searchbind.com         TXT  "v=spf1 include:spf.efwd.registrar-servers.com -all"
_dmarc.searchbind.com  TXT  "v=DMARC1; p=quarantine; rua=mailto:dmarc@searchbind.com; adkim=r; aspf=r; pct=100"

The 61-check run on 7 October scored 93. B2 now passes, the two new email checks pass, and four checks still need work: D2, C9, D4 and D6.

The AI visibility check: 1 of 10

A site score says whether AI engines can read and trust a site. It does not say whether they recommend it. So on 7 October 2026 we ran the same one-time check that the Website report includes: 10 questions a buyer of a GEO audit might ask, each asked once with LLM web search.

The tools recommended most often instead were Otterly.AI (in 3 of 10 answers), Frase (2), HubSpot AEO Grader (2) and Peec AI (2). Those products have been on the market longer and are written about on other sites. A 100 on the site audit does not change that by itself: the audit removes reasons not to cite you, but corroboration from other sites takes time and work.

93 to 100: the last four checks

Four checks were left. We fixed all four the same day, then scanned the live site again.

• C9. The compare page now has a visible byline and a Person author in its Article schema, with the same profile links.
• D4. Three use-case pages, linked from the home page and the footer: for indie SaaS, for local businesses and for online shops.
• D6. This post: a case study with real numbers.
• D2. A "best X for Y" page, one of the 10 high-impact checks: Best GEO tools for solo founders (2026), an honest comparison of eight tools, ours included.

The rescan on 7 October 2026 scored 100/100: 59 checks pass, none fail, and 2 do not apply to our site. A perfect score on our own checklist is a starting line, not a finish: it means AI engines have no technical reason to skip the site. Whether they recommend it is the next question.

What is next

• Earn mentions on other sites. The visibility check shows this is where we are weakest, and no on-site fix replaces it.
• Run the same 10 questions again on our own site once the changes have been live for a few weeks, and publish both numbers side by side.
• Publish customer before and after results, with permission, as they come in.

What this means for your site

A new site can score well on an audit quickly. The fixes that moved our score were small: three links in the schema, two DNS records, a byline, and four honest pages. The harder part is the gap between a good score and being recommended, which depends on what other sites say about you. Fix the site first, because it is fast and it is yours to control, then work on mentions.