We sent a cold outreach email to a fintech founder in April. Standard process — personalised, specific, referenced their product. They opened it. Then, based on what we found when we tested the pattern ourselves, there is a reasonable chance they did something like this before replying:
They pasted our website URL into Claude or ChatGPT and asked: "Is this company credible? Can they build a multi-tenant fintech platform?"
This is happening. We know because we do it ourselves when evaluating vendors. And when we ran that exact evaluation on our own website in April, the results were uncomfortable.
What the AI said about us
Claude's verdict: "Legitimate, early-stage boutique engineering firm worth a scoping call — with caveats."
GPT's verdict: "Credible and technically capable, but not yet proven fintech specialists. Medium-high risk without validation."
Neither recommended us confidently. Both flagged the same gaps: incorporated in 2023 so perceived as young, no third-party reviews on Clutch or G2, no detailed case studies with architecture depth, no explicit answers to the security and compliance questions a fintech CTO would ask.
The uncomfortable part: every one of those gaps was real. Our website was accurate. It just wasn't telling the right story in the right format for an AI to extract and relay to a prospect.
What we learned about how AI evaluates a company
When someone asks Claude or ChatGPT to evaluate a vendor, the AI is trying to answer five specific questions:
What does this company actually do?Who have they worked with and what did they deliver?Are they credible — how long have they been around, who leads it?Are they a fit for my specific problem?What would it be like to work with them?
It answers these by extracting structured meaning from your website content. It doesn't care about your font choice or your motion animations. It cares about clear, specific, factual, unambiguous language.
The problem with most agency websites — including ours — is that they're written for humans who skim, not for AI systems that extract. Vague language gets summarised vaguely. Specific language gets quoted accurately.
The five things we changed
1. We added an llms.txt file
This is an emerging standard — a plain markdown file at /llms.txt that gives AI crawlers a curated summary of your company. Think of it as a briefing document you write specifically for AI assistants evaluating you. No marketing language. Facts, specifics, outcomes.
We wrote ours to answer the exact questions the evaluations flagged: engineering practice since 2007 (not incorporated 2023), the architecture decisions behind our delivered systems, honest answers to the compliance questions fintech prospects ask, and client references available on request.
2. We clarified the company age
Claude read our MCA incorporation date (June 2023) as our founding date. The evaluation came back "three years old, limited track record." The engineering practice started in 2007. We made that impossible to miss — first line of the llms.txt, in the founder section, in the JSON-LD schema.
3. We added structured data that AI can extract
JSON-LD schema markup — Organization, Person, Service schemas — tells AI evaluators exactly who we are, what we do, and how to reach us in a format they can extract with zero ambiguity. Without it, the AI has to infer everything from body copy and can get it wrong.
4. We rewrote case studies to include architecture decisions
The old 7Hub case study said "multi-tenant, RBAC-enforced, with automated underwriting." Accurate but thin. The new one explains why we chose row-level security over schema-per-tenant, why the nightly batch job architecture matches how Italian banks actually work, and why the document builder was built as a no-code tool rather than hardcoded templates.
GPT specifically flagged "no architecture walkthroughs" as a gap. We answered it on the page so the AI can extract it and tell prospects: yes, they explain their decisions.
5. We answered the questions prospects were going to ask anyway
Claude told us: "Probe their approach to security architecture, audit trails, and multi-tenant data isolation — if they answer fluently, that is a strong positive signal."
Claude literally told our prospects what to ask us. So we answered those questions on the services page, in the case studies, and in the llms.txt. Now when a prospect's AI reads our site looking for security architecture depth, it finds explicit answers instead of silence.
What the evaluation says now
We ran the same evaluation after the changes. The language shifted noticeably — from "limited track record, unproven in fintech" to specific references to 7Hub's architecture, the multi-tenant isolation approach, the nightly bank submission system, and the engineering practice predating the entity.
Neither AI recommended against a conversation. Both surfaced the right proof points.
Why this matters beyond our own website
We are not unique in this problem. Every company that sends cold outreach, every agency that relies on inbound, every founder pitching investors — their recipients are increasingly running AI evaluations before responding.
The websites that win are not the ones with the best design. They are the ones that give AI evaluators enough specific, structured, honest information to make a confident recommendation to the human asking.
This is a new layer of the sales funnel that did not exist two years ago. Most companies have not built for it yet.
The practical starting point
If you want to test where you stand: paste your website URL into Claude or ChatGPT with this prompt — "A company reached out to me. Evaluate whether [your company] is credible and a good fit for [your target client's problem]." Read what comes back. The gaps it flags are your priority list.
We shipped our version of this fix at 7unit.tech last weekend. The llms.txt is at 7unit.tech/llms.txt if you want to see what one looks like in practice.