The starting position

FileSeal is a UK-based digital document signing platform. Small team, focused product, competing against DocuSign, Adobe Sign, and HelloSign. When Ben Huss, FileSeal's founder, first ran a SearchScore audit, the results were clear: his site was invisible to AI engines.

The initial GEO score was below 50. Here are the specific issues the audit flagged:

• No structured data at all. Zero schema markup. No Organization, Service, or FAQ schema. AI engines had no machine-readable way to understand what FileSeal did.
• No llms.txt file. The emerging standard for telling AI crawlers what your site is about. FileSeal didn't have one.
• AI crawlers blocked in robots.txt. The default WordPress robots.txt was blocking GPTBot and other AI user agents from crawling the site entirely.
• Generic content. Page copy described the product in the same terms as every other signing platform. Nothing quotable. Nothing an AI engine would cite as a unique answer.
• No entity signals. No Wikipedia or Wikidata presence. No Crunchbase. No third-party verification that FileSeal existed as a real company.
• Weak EEAT signals. No named author. No "about the company" page with verifiable details. No external mentions that an AI engine could use to verify trustworthiness.

The competitive landscape made this worse. DocuSign spends millions on content and SEO. Adobe Sign has Wikipedia pages in twelve languages. HelloSign is backed by a public company. Beating a $40B company in their own category with our budget seemed, frankly, delusional.

What we did

1. Fixed the foundations

First thing: we stopped AI crawlers from being blocked. The default WordPress robots.txt was rejecting GPTBot, Google-Extended, and several other AI user agents. We rewrote it to explicitly allow all major AI crawlers.

Then we deployed structured data across every page. Organization schema on the homepage. Service schema on product pages. FAQ schema on support pages. Article schema on blog posts. Each schema type included the fields AI engines actually read: name, description, URL, sameAs links to social profiles, and aggregateRating where applicable.

We also created an llms.txt file at the root. This is a plain-text file that tells AI crawlers what your site does, what your key pages are, and what your pricing is. It's the equivalent of a robots.txt for LLMs. Most sites still don't have one, which makes it a quick differentiator.

2. Restructured the content for AI extraction

The original homepage copy was a typical SaaS landing page: feature list, button, done. No quotable sentences. No direct answers to questions a buyer would ask an AI engine. We rewrote the key pages to include answer-first paragraphs: self-contained 40-to-60-word blocks that directly answer a specific question.

Example: instead of "FileSeal makes document signing easy," we wrote content that explicitly answered "What's the best alternative to DocuSign for small businesses in the UK?" This is the kind of question ChatGPT gets asked thousands of times per day. If your content doesn't answer it in plain language, you won't be cited.

We also added a dedicated FAQ page with ten questions covering the exact queries we saw in ChatGPT and Perplexity when testing "digital document signing" and "DocuSign alternatives." Each answer was structured as a complete, self-contained paragraph that an AI engine could quote verbatim.

3. Built brand authority signals

AI engines verify that companies exist before recommending them. They cross-reference Wikipedia, Wikidata, Crunchbase, LinkedIn, and other structured sources. FileSeal had almost none of these.

We created a Wikidata entry for FileSeal with the company's core identifiers: name, URL, founding date, industry, and country. This was faster than a Wikipedia article and fed directly into AI knowledge graphs. We added a LinkedIn company page and linked it from the site footer with the correct schema sameAs property.

We also ensured the Organization schema on FileSeal's homepage included comprehensive sameAs links: LinkedIn, X (Twitter), and the company's Crunchbase profile. These entity signals are what separate "a website" from "a real company that an AI engine can confidently recommend."

4. Earned the first AI citations

The breakthrough came about six weeks in. After the foundations were in place (schema, llms.txt, rewritten content, entity signals), we started testing specific queries in ChatGPT and Perplexity every week.

The first citation appeared in Perplexity, not ChatGPT. The query was "alternatives to DocuSign for UK businesses." Perplexity cited FileSeal's FAQ page as a source. That first citation was the signal that the foundations were working. Once Perplexity picked it up, ChatGPT followed within two weeks.

The key trigger, as far as we can tell, was the combination of the llms.txt file and the answer-first FAQ content. AI engines that had previously skipped FileSeal entirely now had a clear, structured reason to index and cite it.

5. Tracked and refined

Once the initial citations appeared, we ran weekly checks across ChatGPT, Perplexity, and Google AI Overviews. We logged which queries cited FileSeal, which didn't, and what competitors appeared instead.

We tried a few things that didn't work. Adding Speakable schema didn't seem to affect citation rates. Publishing more blog posts didn't move the needle as much as improving the FAQ content. The biggest gains came from making existing content more quotable, not from creating more content.

After 90 days, FileSeal was the #1 recommendation in ChatGPT for their core category. The result has held for several weeks since.

The result

FileSeal's SearchScore GEO score moved from below 50 to 75 (Strong tier, top 1% of audited sites). Their technical score reached 98. Structured data hit 100. These aren't vanity metrics; they directly correlate with AI citability.

The ChatGPT result is independently verifiable. You can check it yourself right now:

"We went from invisible in AI search to ChatGPT's #1 recommendation, ahead of DocuSign. SearchScore showed us exactly what was holding us back and then we just executed the fixes. The result speaks for itself."

Ben Huss, Founder, FileSeal

Citations were tracked across multiple AI engines. ChatGPT was the primary win. Perplexity also cites FileSeal for related queries. The result has been stable for several weeks, which suggests it's structural (based on content and entity signals) rather than a temporary ranking fluctuation.

What other founders should take from this

• Beating a giant in AI search costs less than beating them in Google search. The competition is sparser. Most companies, including massive ones, have done zero GEO work. Their content is invisible to AI engines even if it ranks well in traditional search.
• The first AI citation is the hardest. The 50th is automatic if the foundations are right. FileSeal's breakthrough came after six weeks of foundational work, not from a single clever tactic.
• If your content reads well to a human, it might still be invisible to a model. Different problem. AI engines need structured, quotable, answer-first content. Most marketing copy is neither.
• Entity signals matter more than you think. Wikidata, LinkedIn, structured sameAs links. These are how AI engines verify you exist. Without them, you're not a company; you're just a website.
• llms.txt is the lowest-effort, highest-return GEO fix. It takes ten minutes to create. Most of your competitors don't have one. It directly tells AI crawlers what you do.