Grounded AEO logogroundedaeo

← FIELD NOTES

How AI engines choose their sources — and how to become one

When an AI engine builds an answer, it picks sources it can read, verify, and trust. Most businesses fail all three tests without knowing they're being tested. Here's how the tests work — and how to pass them.

Published 2026-07-03 · 7 min read · Grounded AEO

001/THE THREE TESTS

An AI engine answering “who's the best plumber near me?” has seconds to decide who to name. It doesn't deliberate — it filters:

Readable. Can the engine actually parse who you are and what you do?

Verifiable. Do your facts agree with themselves everywhere they appear?

Trustworthy. Is there proof the facts are current and maintained — not scraped from 2019?

Fail one test and you drop down the list. Fail all three and you simply don't exist to the engine — no matter how good your work is.

002/TEST ONE — READABLE

A beautiful website and a readable website are different things. Engines parse structure, not design: real server-rendered pages, structured facts (your name, address, offering marked up so machines can extract them), and a machine-readable guide at your site's root.

There's also a door test most owners never think about: your robots.txt file can block AI crawlers entirely — sometimes by accident, from a rule added years ago. If the crawler can't come in, nothing else on this list matters.

003/TEST TWO — VERIFIABLE

Engines cross-check. If your website says one address, an old directory says another, and your social profile says a third — the engine doesn't pick one. It concludes your facts can't be trusted and moves on to a business whose facts agree.

This is the single most common failure for local businesses: the name/address/phone drift that accumulates over years of moves, rebrands, and forgotten listings. Consistency is boring, unglamorous — and load-bearing.

004/TEST THREE — TRUSTWORTHY

Two sources can both be readable and internally consistent — and one still wins: the one with provenance. Facts with a visible maintenance history (published dates, refresh timestamps, a verified record proving domain ownership) beat facts that haven't moved in five years.

Trust signals also compound with tenure — the way domain age became a ranking signal in Google. A record verified and refreshed for years outweighs an identical record created yesterday. Which is the honest argument for starting now rather than waiting for AI recommendations to feel urgent: by then, the tenure belongs to someone else.

005/THE PHONE-BOOK PROBLEM

Here's what happens in the many niches where nobodypasses the tests: the engine still has to answer. So it assembles one from whatever scraps it can find — stale directories, old reviews, half-parsed pages. The internet's version of an old phone book. Nothing verified, and nothing in the business's own words.

That's the real opportunity. In a niche where AI is answering from the phone book, the firstbusiness to hand it something readable, verifiable, and maintained doesn't just improve its odds — it becomes the answer by default. Most local niches are still in exactly this state.

006/QUESTIONS, ANSWERED
Why does AI cite my competitor and not me?+

Usually not because they are better — because their information is in a form the engine can read, verify, and trust, and yours is not yet. The engine is not judging quality; it is judging verifiability. That is fixable.

Do good reviews make AI recommend me?+

They help, but they are one signal among several — and reviews live on other platforms, in other people's words. The signals you fully control are your own: structured facts, a machine-readable guide, consistency, and a verified record.

Is this different for ChatGPT vs Google vs Perplexity?+

The details differ; the three tests do not. Every engine needs to read, verify, and trust a source before naming it. Pass the tests once — on your own domain — and you are positioned across all of them.

How long does it take to become a cited source?+

Nobody can honestly promise a timeline. What is measurable: whether you currently appear in AI answers for the questions that matter to you. Measure first, fix the failing tests, then watch the number — that is exactly what a citation dashboard is for.

What single thing should I do first?+

Find out where you stand — the failing test is different for every business. A free audit shows which of the three tests you fail and in what order to fix them. Guessing wastes months; measuring takes 30 seconds.

Which test do you fail?

The free audit runs all three against your domain and shows every gap in plain English. 30 seconds, no account.