Ask an AI assistant for "a good accountant near Rome, NY" and it will name some firms. The owners of those firms did nothing special to be chosen — and the owners of the firms that weren't named have no idea they're invisible. We build and operate an audit product that checks exactly this, so we've looked at how real businesses fare in these answers many times over. The mechanics are less mysterious than the market wants you to believe.
Where the answers come from
Modern assistants answer local and commercial questions by reading the live web — searching, fetching pages, and synthesizing what they find, sometimes blended with what their training data already contains. Both paths run through your public presence: your website, your Google Business Profile, directories, reviews, and every place your name, address, and phone number appear.
This means the question "how do I rank on ChatGPT?" is mostly the old question — "is my business legible to machines?" — asked with higher stakes, because an AI answer names three businesses, not ten blue links.
What the machines actually reward
Running audits across real businesses, the same handful of factors decide most outcomes:
- Structured data. Schema markup is the difference between a machine parsing "we're open till 5" from a paragraph and knowing your hours as data. Businesses with correct LocalBusiness/Service schema get represented accurately; businesses without it get guessed at. (Plain-English explainer: the dictionary.)
- Consistency. If your name, address, phone, and hours disagree across your site, Google profile, and directories, a machine reconciling sources treats you as uncertain — and uncertainty loses to competitors who are boringly consistent.
- Legibility. Assistants fetch pages under time pressure. A fast page whose actual content is in the HTML gets read; a slow page that assembles itself with scripts often effectively doesn't exist. Speed stopped being cosmetic the moment machines became your most impatient reader.
- Answers to real questions. Pages that plainly state what you do, where, for whom, and for how much give the machine sentences it can safely repeat. Vague brochure copy gives it nothing to quote.
- Corroboration. Reviews and third-party mentions function as evidence. An assistant hedges about a business only its own website describes.
What doesn't work
The tricks are already dead on arrival. Keyword stuffing reads worse to a language model than to a human. "AI optimization" services promising secret access to model rankings are selling something that doesn't exist. And a fake-review strategy now fails in front of software that's rather good at spotting patterns. The uncomfortable, liberating truth: the machines reward exactly what a careful human visitor always did — clarity, consistency, speed, and evidence — enforced with less patience.
What to actually do
- Check what the machines see. Ask a few assistants about your business and your category. Run an audit — ours or anyone's honest one.
- Fix the structure: schema markup, llms.txt, meta descriptions, one consistent name/address/phone everywhere.
- Fix the speed: images, bloat, and whatever your page builder bolted on.
- Write pages that answer the questions people actually ask, in plain sentences.
- Re-check. This is the part most providers skip: the same audit, re-run, should show the score moving. If improvement isn't measurable, it didn't happen.
None of this is exotic. Most of it is a fix list, not a retainer — which is why our packaged version, the $499 Fix Sprint, is a one-time job with a before-and-after score, not a subscription. And if your presence is already in good shape, the honest output of an audit is: you're fine, spend your money elsewhere.