When an AI engine decides whether to recommend a local business, it isn't running one calculation. It's quietly answering four separate questions. The REAL Method splits your AI Recommendation Rate into those four questions so a single low number turns into a specific, fixable to-do list.
Here's the short version, then the detail. REAL stands for Recognize, Evidence, Answer, Link. Each layer is scored 0–25. Add them up and you get a 0–100 recommendation rate. To be clear about what this is: it's our scorecard, not an official ranking from Google, OpenAI, Anthropic, or Perplexity — none of them publish one. It's a working model that maps to how answer engines actually behave.
R — Recognize: does the model know you exist?
Recognize is the entity layer. Before an engine can recommend you, it has to be confident that you are a specific, real local business — not a guess, not a confusion with a similarly-named company in another city.
This is where most invisible businesses are actually failing. The model isn't rejecting you; it doesn't have a clean enough picture of you to risk putting your name in an answer. What this layer checks:
- LocalBusiness schema with a specific subtype (
Dentist,Plumber,RoofingContractor), not genericOrganization. - NAP consistency — name, address, phone identical across your site, Google Business Profile, and directories. Conflicting details confuse entity resolution.
sameAslinks connecting you to authoritative profiles (Google Business Profile, Wikidata, social accounts) so the engine can verify you're one real entity.- A clear About page that states who you are, what you do, and where you serve in plain language.
A strong Recognize score means every major engine can describe what you do and where, correctly, on a direct name query. A weak one means one engine nails it and the others give a generic answer that could describe any business in your category.
E — Evidence: is there enough proof to back a recommendation?
Recognize asks "do they know you?" Evidence asks "is there a reason to pick you?" Engines don't trust you about you — they trust other people about you, plus the specifics on your own pages.
This layer weighs third-party and proof signals: review volume and recency, named mentions, real service detail, and concrete pricing. It matters more than people expect because the research keeps pointing at it. The widely-cited Princeton/Georgia Tech GEO study (KDD 2024) found that adding statistics and cited sources to content were among the most effective ways to increase how often generative engines surfaced it — while keyword stuffing actually reduced visibility. Vague marketing language is the opposite of evidence.
A practical read: ten Google reviews and nothing else is a fragile position. Reviews across multiple platforms, recent dates, and pages that state actual numbers (timelines, price ranges, what's included) is what an engine can stand on when it recommends you.
A — Answer: can your pages be quoted verbatim?
The Answer layer asks whether a page on your site directly answers the questions customers actually ask — in words the engine can lift straight into its reply.
This is the most mechanical layer to improve and one of the highest-leverage. Answer engines reward self-contained, extractable passages. Analyses of cited content repeatedly find that the passages engines pull tend to answer the core question early and completely: Google's AI Overviews favour compact, self-contained passages, and studies of Perplexity citations find the large majority answer the core question within the first hundred words or so.
In practice, Answer Coverage is the percentage of your real customer questions that have a clean, quotable answer somewhere on your site:
- A cost guide with real ranges, not "contact us for a quote."
- An FAQ that answers permit timelines, what's included, how long jobs take — in 40–80 word chunks.
- Service pages that lead with the answer instead of burying it under brand copy.
The customer question you don't answer on your site is the one a competitor's page will answer for them — and that competitor gets named instead of you.
L — Link: can AI crawlers actually reach you?
Link is crawlability, and it's the floor under everything else. You can win the other three layers and still get zero citations if the bots can't fetch and parse your pages.
This is also where the old advice is dangerously wrong, so it's worth being precise. The crawler that controls whether ChatGPT can cite you in search is OAI-SearchBot — not GPTBot, which is only for model training. Blocking GPTBot does not remove you from ChatGPT search; blocking OAI-SearchBot does. Anthropic works the same way: Claude-SearchBot governs citations, ClaudeBot is training. Google AI Overviews ride on regular Googlebot, not Google-Extended. The Link layer checks:
- Whether the citation-grade crawlers can reach your key pages.
- Whether content renders in raw HTML or hides behind client-side JavaScript that crawlers skip.
- Whether your schema validates and your sitemap includes the pages you want cited.
I once audited a site with excellent reviews and real authority whose hero content rendered entirely client-side — every crawler was reading a blank page. The other three layers didn't matter until that was fixed.
What the composite number is actually for
The overall AI Recommendation Rate is the headline, but the work lives in the layers. When I open a new account, I look at the lowest layer first — it's almost always the one holding the other three back.
A simple way to read your four numbers:
- 20–25 on a layer: strong. You're defending the position, not building it.
- 12–19: mixed. There are fixes here that move the number in two to four weeks.
- Under 12: invisible on that layer. Not a judgement — a starting point. Most sites start here on at least one layer.
You can see your own four REAL numbers, per engine, in about 90 seconds with the free AI Visibility check — no signup. If you want the weekly re-scoring, competitor gaps, and Flare walking you through fixes in priority order, the plans live at /pricing/. And if you want to go deeper on any single layer, the Robots Check and GEO Grader each isolate one of them.