From “I have no idea” to a scored answer in minutes

Where the buyer searches come from

The quality of an AEO audit is decided before a single engine is queried. Run the wrong questions and you get a confident report about searches nobody makes.

We build the search set from how customers actually phrase a request to an assistant, which is different from how they type into a search box. Search-box behaviour is terse and keyword-shaped: "hvac columbus". Assistant behaviour is conversational and constraint-loaded: "my furnace is making a grinding noise and it is 20 degrees out, who in Columbus can come today and roughly what will it cost?"

So the set spans the whole decision, not just the bottom of it — problem-framing questions, comparison questions, price and trust questions, and the direct "who should I use" questions. It includes your category, your service area, and the specific services you say you offer, so a gap in one line of business does not get averaged away by strength in another.

How the answers are scored

Each answer is read for four things, and the Findability Score is the weighted result. Being named in the answer counts most, because that is the outcome that produces a customer. Being cited as a source counts next, because citation is durable in a way a passing mention is not.

Breadth across engines counts, because appearing only in Gemini usually means you are visible through your Google Business Profile rather than through anything on your own site — real coverage shows up in more than one model. And accuracy counts, because being named with the wrong phone number is worse than not being named.

The score is always shown with its components. A number you cannot decompose is a vanity metric, and we would rather hand you a 34 you understand than an 82 you cannot act on.

Why the same question gives different answers

Answer engines are non-deterministic. Ask the same question twice and the wording changes; ask it next week and the recommendation may change. This is the single most misunderstood thing about the category, and it is why a report is a measurement rather than a guarantee.

We handle it two ways. Within a run, the search set is large enough that a single unlucky answer cannot swing the score. Across runs, the membership re-tests the same searches so you are comparing like with like — the trend is the signal, and one report is a starting point rather than a verdict.

What you do with the results

The fix plan is ordered by how much of the visibility gap each item closes, and almost every plan starts in the same place: make your facts machine-readable. Publishing an llms.txt, adding Organization and Service schema, and moving critical facts out of PDFs and images into text closes more ground than anything else, because most businesses fail not on authority but on legibility.

After that comes correcting the stale facts the engines picked up from third parties, then building answer pages for the specific questions where a competitor is currently named instead of you. If you would rather have the files than the instructions, the Developer Bundle produces them.