← All essays
- AI Search
9 min read · July 2026
The Proof Layer: reviews are infrastructure now
Every answer engine reads the human trust signal before it decides whether to name you. Reviews stopped being a vanity metric the moment machines started quoting them.
Sudhir Sharma
ex-Google · ex-Adobe · advisor
For twenty years, reviews were treated as a reputation problem — something the marketing team monitored, occasionally panicked about, and mostly hoped would take care of itself. That framing is now dangerously out of date. Reviews aren’t reputation anymore. They’re infrastructure — a load-bearing part of how you get discovered at all.
Here’s what changed. When a customer asks an AI assistant for “the best physiotherapist near me” or “a reliable fractional CFO for a Series A,” the model doesn’t invent an answer. It assembles one — from structured data it can read, and from the human signals it can find about who’s trustworthy. Your star rating, the recency of your reviews, whether you reply to them, what specific words customers use: all of it is now raw material the machine reads before it decides whether to put your name in the answer.
The algorithm decides if you’re eligible to be named. The reviews decide if you deserve it.
Why "infrastructure" is the right word
Infrastructure is the stuff you only notice when it fails. Nobody thanks the plumbing; everybody notices the flood. Reviews now behave the same way. When they’re strong, fresh, and answered, they quietly do their job in the background — feeding the algorithm the confidence it needs to surface you. When they’re thin, stale, or ignored, you don’t get a warning. You just quietly stop appearing, and you never find out why.
This is the trap I see most often in a teardown: a business that’s genuinely excellent, with delighted customers, that has treated reviews as something that “happens” rather than something they build. The excellence is real. But it’s unstructured — trapped in customers’ heads and private WhatsApp recommendations, where no machine can read it. In the AI era, unstructured excellence is unciteable excellence.
- The shift in one line
Reviews used to be about protecting your reputation with the humans who found you. Now they’re about being found in the first place — because the machine reads them before the human ever sees you.
What actually feeds the machine
Not all review signals are equal. When I audit a business’s proof layer, four things matter far more than the raw star number:
- Recency. A five-star average built three years ago reads as a business that was good. Engines weight fresh signal heavily — a steady trickle beats a historic pile.
- Response. Replying to reviews — especially the critical ones — is a signal of a live, accountable business. It also gives the machine more text, in your voice, to read.
- Specificity. "Great service!" tells a machine nothing. "Fixed my knee after two other clinics couldn't" tells it exactly what to name you for. The language in your reviews is what maps you to a query.
- Spread. One platform is a single point of failure. Presence across the sources an engine actually reads — Google, industry directories, the places your buyers check — is what makes the signal robust.
The move
Treating reviews as infrastructure means building a system, not running a campaign. A light, repeatable routine that asks for reviews at the right moment, makes leaving one frictionless, replies to every one, and steers happy customers toward specific language about what you actually did for them. It’s not glamorous. Infrastructure never is. But it’s the difference between being the business AI names and the business it forgets.
If you want to know where your proof layer stands today — strong and feeding the machine, or quietly costing you the answer — that’s exactly what the diagnostic measures on the Ego axis.
Where does your proof layer stand?
The AI Visibility Diagnostic scores your trust signal against your visibility — ten questions, three minutes, your position on the map.