Google’s ecosystem
- Google AI Overviews
- Gemini
Reads the Google Business Profile and Google reviews. An agent with a thin GBP is underweighted here no matter how strong the open web looks.
Worked sample · agent vertical
An agent audit,
worked end to end.
This is what the AuthorityMap agent audit looks like on the inside: the question set, the engine split, the scorecard, and the compliance gate — demonstrated on a fictional Eastside WA agent so no real client’s data is exposed.
1 · The question set
Agents are found by place, not program.
A loan officer is asked about programs. An agent is asked about places. So the agent prompt set names specific neighborhoods and towns — because that’s what buyers type — and measures whether the agent (or the agent’s neighborhood page) is the source engines reach for.
| Buyer’s question | What we measure | Sample baseline demo |
|---|---|---|
| “best real estate agent in Maplewood Heights Bellevue” | Is the agent named, and on what evidence? | Absent on every engine tested demo |
| “who are the top agents for Lake Hills neighborhood” | Does the agent enter the consideration set? | Absent demo |
| “homes for sale in Maplewood Heights” | Is the agent’s neighborhood page cited as a source? | Portals cited; no individual agent named demo |
Two disciplines apply to every prompt. Sampling: each question runs at least 7 times per engine, so a “named” or “absent” result carries a confidence interval — one AI answer is an anecdote. Steering-safe design: we never author prompts that bait an answer about who lives somewhere. Where a buyer’s natural phrasing implies one, we observe what engines say but reframe every finding and every recommendation to place facts — prices, days on market, walkability scores, and other published city data — never occupant types.
2 · The engine split
Two camps read two different records.
The five engines don’t read the same sources, so a fix that moves one camp can be invisible to the other. The audit routes every recommendation to the camp that can actually see it.
Open-web engines
- ChatGPT
- Perplexity
- Claude
Cannot see Google reviews at all. They read Zillow, Redfin, Yelp, press, and the agent’s own pages — the open-web record decides these answers.
3 · The scorecard
Five axes, built for how agents get cited.
The agent scorecard swaps the mortgage vertical’s program-readiness axis for neighborhood-page readiness: does the agent own a citable, numbers-first page for each target neighborhood?
| Axis | Sample score | What it means here |
|---|---|---|
| Entity consistency | 3 / 5 demo | State-license link present; profile names drift across surfaces. |
| Neighborhood-page readiness | 1 / 5 demo | No numbers-first place page yet — the single highest-leverage fix. |
| Off-site citation presence | 2 / 5 demo | Thin on the open-web surfaces the non-Google engines read. |
| Review aggregate (open web, dated) | 3 / 5 demo | Present on one portal; thin elsewhere. Counts treated as directional. |
| Schema (Person + RealEstateAgent + Place) | 2 / 5 demo | No Place node on neighborhood pages; license link is the identity anchor. |
Every score is synthetic here — the point is the shape. In a real audit each score traces to captured engine answers and public-record checks, and each becomes ranked items in the 30-day fix queue.
4 · The compliance gate
The gate catches what SEO copy habits get wrong.
We wrote two versions of a realistic agent bio and neighborhood page for the fictional Dana Reyes, then ran both through our real-estate compliance scanner. The difference is the product.
Draft · refused to ship · 30 blocking flags demo
The “before”: typical hyperlocal marketing copy.
Written the way agent copy often reads in the wild. The scanner refused it with 30 blocking flags demo across three rulesets (phrases redacted here — reproducing steering language on a public page is itself a violation):
- Fair Housing: wording that signals who a home or neighborhood is “for” — touching familial status, religion, race, national origin, disability, source of income, and age.
- Post-settlement practice rules: how the cost of buyer representation was framed, and where payment terms were surfaced.
- REALTOR® mark: wrong casing, possessive use, and the mark attached to a team name.
Rewrite · passed clean · 0 flags demo
The “after”: the same page, written with numbers.
The rewrite says what buyers actually need to know — about the place, with published numbers and named sources — and passed the same scanner with zero flags demo:
- Describes the place with data: median price, days on market, Walk Score, and other published, dated city statistics.
- States the written buyer agreement and keeps compensation off the MLS.
- Uses the REALTOR® mark exactly as the mark’s rules require.
“Maplewood Heights has a median sale price of $612,000, and homes spend a median of 18 days on market. Walk Score is 64. … Reported property crime is 11 per 1,000 residents per the city’s 2025 open-data report.” demo
— from the passing rewrite. Fictional numbers, real technique: describe the place with numbers and named sources, never the occupant.
One honest boundary: the scanner is a high-precision first pass, not a replacement for review. It exists so every draft in a fix queue arrives pre-screened, and a human compliance read stays fast and consistent.
Your market, real answers
This was the fictional version. Yours would be real.
Same question design, same engine split, same scorecard, same gate — run against what AI actually says in your market.