How the Bizer Directory makes small businesses discoverable, and answerable, by ChatGPT, Google AI, and the agents your customers now ask instead of searching. A technical deep-dive: AI-facing business profiles, an agent that reads your entire website, structured data by construction, and honest AI-visibility measurement.
The AI-discovery shift: your next customer is asking an AI, not a search box
Ask a question like “who’s the best veteran-owned butcher near Cut Off, Louisiana?” and increasingly you don’t scroll ten blue links. An assistant just answers, naming a few businesses. That answer is the new storefront window. If an AI can’t read a business, understand what it does, and trust its facts, the business simply isn’t in the answer.
This is the shift behind Answer Engine Optimization (AEO) and Generative Engine Optimization (GEO): being cited by AI systems the way you once optimized to rank on Google. For a national brand with an SEO team, that’s a new workstream. For the plumber, the meat market, and the mobile-detailing shop that Bizer serves, it’s a workstream that will never happen. Most small businesses will never hand-craft structured data or study how a language model describes them.
“Most small businesses will never do AEO. So the Bizer Directory does it for them, by construction.”
That sentence is the whole thesis of the Bizer Directory. This article is a look under the hood at how we make a business found by AI on the outside while the owner keeps running it, with AI on the inside. None of it relies on tricks or invented metrics, which, as you’ll see, is itself a design decision.
The problem: modern web pages serve AI nothing
Most business listings today are single-page apps. A crawler that fetches the URL gets an empty <div id="root"> and a bundle of JavaScript it won’t execute. The exact pages we most want an AI to read serve it a blank shell. Meanwhile the facts that would earn trust (who owns it, what it sells, whether it’s a real, registered, active business) are scattered across a website, a registry, and the owner’s head.
We also tested the popular shortcut and set the record straight internally: an llms.txt file is not the mechanism. It’s widely published and, by the evidence, widely ignored by the AI systems it targets. What AI models actually learn from are crawlable, structured, factual pages. So that’s what we build.
The Bizer Profile: AEO by construction
Every business in the Bizer Directory gets a Bizer Profile at a permanent, canonical URL (/biz/{state}/{slug}). Behind that one address we render the same facts three ways, so every kind of reader is served first-class:
- Server-rendered HTML. The profile’s content is injected into the page on the server, so a crawler (or a human with JavaScript off) reads a real page, not an empty shell. Identical HTML for people and machines: no cloaking.
- A Markdown twin at
/biz/{state}/{slug}.md, the clean, structured format AI crawlers prefer, with the facts and their sources laid out plainly. - Structured data (JSON-LD).
LocalBusiness,FAQPage, and a Recent-jobsItemList: the machine-readable layer an assistant parses to know exactly what a business is, offers, and has done lately.
Crucially, all of this is generated from the record, never invented: the state Secretary of State registry, the business’s own website, and what the owner has explicitly published. A Bizer Profile is trusted by the sources behind it, not by star ratings. That’s the anti-fabrication spine that runs through everything below.
Miles: an agent that reads your whole website to answer
A profile isn’t a brochure. It’s talkable. Every listing has Miles, the business’s AI assistant, and a visitor can ask it anything. The interesting part is how Miles answers.
The naive approach (the one we started with and then tore out) is to crawl a site once, have a model extract it into a fixed set of fields, and answer from those fields forever. It’s lossy: anything the schema didn’t think to ask for is gone. Ask “who owns this business?” and you get “veteran-owned” (a captured trait) but not the owners’ names, even though they’re right there on the About page.
“Miles is an agent. When a question needs an answer, it reads the live website to find it.”
So Miles reads the business’s own site on demand: it opens the homepage, follows the links it finds (About, Our Team, Menu, Services) and keeps reading until it has the answer, then replies in its own words. No fixed page limit. The one hard rule never bends: if the site doesn’t say it, Miles doesn’t say it. It answers from the public record, the owner’s published profile, or a page it actually read this turn, and nothing else. For safety, the read tool is locked to the business’s own domain, so it can never be steered to fetch somewhere it shouldn’t.
The whole-site brain: read everything once, remember all of it
On-demand reading is great for freshness but pays a round-trip per question. So the moment a business has a website on its profile, Bizer builds its brain: a background job crawls the entire same-origin site, every page, breadth-first, and stores it as the business’s standing knowledge. From then on, Miles has instant, total recall of the whole website, and the live-read tool is just there for anything genuinely new.
It’s bounded only by sane safety guards (a page ceiling, a size ceiling, a time budget) that a normal small-business site sits comfortably under, not by a knowledge-shaped cap. And it keeps itself current: the brain rebuilds when a site changes, and a nightly job refreshes on a cadence. In effect, Bizer becomes the business’s memory.
Freshness the algorithms reward: Recent jobs, as data
Search and AI both reward fresh, local activity. So an owner can post a one-line update after a job, like “fresh andouille batch is in the case,” and Miles turns it into a dated, place-tagged entry on the profile. These Recent jobs render for human visitors and, just as importantly, ship as a structured ItemList in JSON-LD so an assistant can answer “has anyone done a job like mine lately?” Privacy is structural, not a setting: there is no customer field, and locations are place-level (a town), never an address.
Measuring AI visibility, the honest way
Here’s the part we’re proudest of, precisely because of what it refuses to do. It would be easy to slap a made-up “AI Visibility: 87%” gauge on a dashboard. We don’t. Instead, we measure it.
For a claimed business, Bizer builds real local questions from its own category, place, and offerings (the queries a customer would actually type) and asks them to real engines: ChatGPT (OpenAI) and Google AI (Gemini). Each engine is queried directly, so a “Google AI” result is really Gemini and a “ChatGPT” one is really OpenAI. Then a deliberately strict detector checks whether the business was genuinely named in the answer: a real name, not a generic category word. The score is simply the share of answers that named you.
The owner sees the evidence, not just a number: the actual question, whether each engine named them, and the real snippet of what it said, alongside the true count of AI crawlers that fetched their page in the last 30 days. It runs on demand and refreshes nightly, so the number trends over time. When we can’t measure something truthfully, we don’t show it: there’s no Perplexity score because we don’t query it, and Claude joins the panel the day its key is added, not before.
“Facts with provenance, never an estimated score. If we can’t measure it, we don’t show it.”
How it fits together
- Canonical profile. One permanent URL per business, rendered as HTML plus a Markdown twin plus JSON-LD, first-class for people, crawlers, and assistants alike.
- Grounded facts. State registry, the business’s own website, and owner-published details. Every fact carries a source; nothing is invented.
- Talkable agent. Miles answers from the live website, the whole-site brain, and the public record, and says “I don’t know” rather than guess.
- Whole-site brain. The entire website read once and kept current, so the agent has total recall without a fetch per question.
- Structured freshness. Owner updates become dated, place-tagged “Recent jobs,” emitted as JSON-LD an assistant can query.
- Measured visibility. Real probes to ChatGPT and Google AI, strict mention detection, and real crawler-read counts: measured, never estimated.
Every layer shares one discipline: say only what a source supports.
Frequently asked questions
What is Answer Engine Optimization (AEO)?
AEO is making your business legible and trustworthy to AI answer engines (ChatGPT, Google AI, and similar) so they name and describe you accurately when a customer asks. It’s the successor discipline to SEO for a world where an assistant answers instead of listing links.
How does Bizer help a business get found by AI?
Bizer gives every business a Bizer Profile that’s crawlable by construction: server-rendered HTML, a Markdown twin, and structured JSON-LD, all built from verified sources. That is exactly the form AI systems read and trust, so the business can be named in AI answers without the owner doing any technical SEO work.
Does Bizer invent facts or ratings to look good?
No. Every fact on a profile is grounded in the state registry, the business’s own website, or what the owner published, and AI visibility is measured by really asking the engines, not estimated. If something can’t be verified or measured, Bizer doesn’t show it.
Which AI engines does Bizer measure visibility on?
Today, ChatGPT (OpenAI) and Google AI (Gemini), each queried directly. Additional engines are added only when they can be measured truthfully, never faked to fill a chart.
Bizer is the business OS you scroll: run with AI on the inside, found by AI on the outside. Claim your Bizer Profile and see how AI describes your business.