How to get your business recommended by ChatGPT, Perplexity and Google's AI Overviews. The owner's playbook.

Three years ago a buyer typed "estate lawyer syracuse" into Google, scanned ten blue links and three map pins, and clicked a couple. Now a growing share of them type "my mother died without a will, which estate lawyer in Syracuse should I call" into ChatGPT, or get an AI Overview at the top of Google that answers before the links appear, and they get a shortlist: three or four names, a sentence on each, and the sources. The shortlist is the new first page, and it's shorter.

I spend most weeks inside the websites, listings and ad accounts of owner-led companies, and the question I'm asked most now is some version of "how do we get on that list?" This is the full answer, as I'd give it to a client: how the engines build the list, the four things they need to find before they'll name you, the check to run today, the plan for the next thirty days, how to measure it, and the things that don't work, with Google's own words where it has given them.

How the engines build the shortlist

The engines differ in what they read, and that matters for where you do the work.

What they share: for a recommendation question, they retrieve pages about businesses that fit, extract the facts they can support, and name the ones they're confident about. Confidence comes from agreement between sources. A business that states its facts clearly, and that other sources agree with, is nameable. A business whose own pages disagree with each other, or that nobody else mentions, gets left off, not out of malice but because the engine would rather name the competitor it's sure about.

The four things an engine needs before it names you

1. Identity: one set of facts, everywhere

Name (exact spelling), legal name, founding year, locations, phone, hours, service area, the people and their credentials, the services and what they cost, the review count by platform with a date. Written once on a one-page fact sheet, then applied to every page, listing and piece of markup. The most common reason a well-regarded business isn't named is that its own website carries three review counts and two founding years, and its directory listings carry a fourth and a third. The fact sheet template is here.

2. Evidence: other sources that say what you say

The engines cite third parties for "who should I hire" far more than they cite the business itself: Google reviews, the review and directory platforms for your category, the regulator's licence database, manufacturer or association listings, local press, and community threads. If the only place on the web that says you're excellent is your homepage, there's nothing to corroborate. The prompt check below tells you exactly which sources the engines trust for your category in your market, so you build presence where it counts instead of everywhere.

3. Quotable pages: sentences a model can lift without editing

Open your main service page and try to answer, from the text alone: what exactly do you do, where, for whom, since when, and for how much? If the answer is a paragraph about passion and excellence, the engine has nothing to quote. "Flat-fee estate plans from $1,800, two meetings, signed within three weeks, Onondaga and Madison counties since 2009" is quotable, and quotable wins. Google's guidance puts it as content that brings "real experience" rather than summarizing what already exists, and warns that content that "could easily be produced by a generative AI model" adds little. Specific, first-hand, numbered sentences are the opposite of that.

4. Access: the crawlers can reach you

The page has to be indexed and eligible to be shown with a snippet (Google's stated requirement for its AI features), publicly accessible, and not blocked to the crawlers that feed the answers. Two things to check in the first hour: that the site is verified in Bing Webmaster Tools and Search Console with a sitemap submitted, and that robots.txt doesn't block the search crawlers. Here's a robots.txt that allows the answer engines while leaving your training-crawler decision to you:

# Allow the search/answer crawlers you want citing you.
# (Blocking them removes you from the answers, not from the training sets you may object to.)
User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Bingbot
Allow: /

# Google-Extended controls Gemini training/grounding only.
# AI Overviews and AI Mode use the regular Googlebot, so this line does not affect them.
User-agent: Google-Extended
Allow: /

One distinction worth knowing: Google-Extended controls whether your content is used for Gemini training and grounding. It does not affect AI Overviews or AI Mode, which use the ordinary Googlebot. Blocking it won't remove you from Google's AI answers, and allowing it won't add you.

The ten-prompt check (run it today)

Write ten prompts the way a buyer would type them, not as keywords: "who are the best X in [place]," "I need X for [situation], who should I call," "how much does X cost in [place] and who's reputable," "compare [you] and [competitor]," "is [you] any good." Fill the rest from your last ten sales calls. Run each on ChatGPT with search on, Perplexity, Gemini, and a plain Google search to see whether an AI Overview appears. Log it like this:

PROMPT (as a buyer would type it)        | ENGINE      | NAMED? | WHO WAS NAMED              | SOURCES CITED                         | NOTE
Best estate lawyers in Syracuse NY       | ChatGPT     | no     | Firm A, Firm B, Firm C     | avvo.com, firm B site, reddit         | Firm B page states flat fee
Best estate lawyers in Syracuse NY       | Perplexity  | yes    | Us, Firm A, Firm B         | google reviews, our /estate-planning/ | cited our price sentence
Best estate lawyers in Syracuse NY       | Gemini      | no     | Firm A, Firm D             | firm A site, superlawyers              |
Best estate lawyers in Syracuse NY       | Google AIO  | no AIO | (map pack shown)           |                                       | AI Overview not triggered
How much does a will cost in NY          | ChatGPT     | n/a    | (answered with a range)    | nolo.com, firm B site                 | we have no cost page

Forty rows. Three columns matter: whether you were named, who was named instead, and the sources. The sources column is the diagnosis and the plan: it tells you which pages, platforms and directories the engines trust for your category, and it usually shows you the exact sentence on a competitor's site that got them named. (In the example above, the firm has no cost page and the engine answered the cost question from a competitor and a legal-information site. That's a page to write this week.)

Keep the spreadsheet. Re-run it monthly. It's the only AI-visibility report that matters, and it's the one I send clients. The twenty-minute version with more prompt examples is here; the industry versions for law firms and home services have the prompts for those categories.

The 30-day plan

After day thirty: two to four specific pages a month written from real experience and signed by a real person, listings and reviews maintained as a system, the prompt log re-run monthly, and a quarterly look at whether the paid account is still buying searches the organic side now wins. That last item usually pays for the rest.

What kind of content gets cited

Not more content. More specific content. The pages I see cited most, across categories:

• Cost pages with ranges, assumptions and a date. "How much does X cost in [place]" is asked constantly and answered from whoever says.
• Process pages that walk through what actually happens, step by step, in your jurisdiction or your trade, with timeframes.
• Problem pages for when something has gone wrong: denied, rejected, delayed, failed, revoked. Urgent intent, almost no competition.
• Comparison pages that honestly lay out you versus the alternative (including "do it yourself" and "do nothing"), because the engines are asked to compare and will cite whoever did the work.
• Project and case pages with real places, numbers and outcomes, under a real byline.
• Video with a transcript. The engines cite YouTube; a transcript on the page makes the same content quotable.

And the thing that gets ignored: the "ultimate guide" that could have been written by anyone, about anywhere. Google's guidance says it in so many words, and the other engines behave the same way.

How to measure it

The prompt log

Monthly, same prompts, same engines, same spreadsheet. Track three numbers: prompts where you're named (out of forty), prompts where you're named first, and the count of distinct sources citing you. That's the report.

Search Console

Since June 2025, clicks and impressions from AI Mode have counted in the Performance report's Web totals, folded in rather than separated. In June 2026 Google began rolling out a Generative AI performance report (to a subset of sites first) that shows how often your URLs appear in AI Overviews and AI Mode, by page, country, device and date. If your property has it, it's the closest thing to an official AI-visibility number for Google; if not yet, the folded-in totals and the prompt log carry you.

GA4: traffic from the chat engines

Referrals from ChatGPT, Perplexity, Copilot, Claude and the rest show up in GA4 with their hostnames as the session source. A free-form exploration with a regex filter separates them:

GA4 → Explore → Free form
Dimension: Session source (or Page referrer)
Metric: Sessions, Key events
Filter: Session source  matches regex:
chatgpt|openai|perplexity|copilot|gemini\.google|claude\.ai|you\.com

Notes:
- Google AI Overviews and AI Mode arrive as google / organic. They cannot be separated here.
  Use Search Console's Generative AI performance report for those.
- Some engines strip the referrer; treat this as a floor, not a total.

Treat the number as a floor. Some engines strip the referrer, and Google's AI features arrive as ordinary organic traffic. The point of the report isn't the total; it's seeing which pages the chat engines send people to, because those are the pages they're citing.

Leads, by page

Call tracking with dynamic numbers and booking events passed to analytics, so that organic and AI-referred leads are counted by the page that produced them. Without this, "AI visibility" is a vanity metric. With it, it's a line in the same report as the paid account, and it can earn budget on the same terms.

What doesn't work (with Google's words)

• llms.txt. Google: "Google Search itself doesn't use them." Harmless, and not the lever.
• Chunking and rewriting for AI. Google says there's "no ideal page length" and no need for special phrasing, since the systems understand synonyms and intent. Write for the reader.
• Manufactured mentions. Google says seeking inauthentic brand mentions "isn't as helpful as it might seem." A burst of thin directory listings also tends to introduce new inconsistencies, which is the opposite of what you need.
• A page for every sub-question. Generating pages at scale "mainly to manipulate rankings" is called out under the scaled content abuse policy. Five real pages beat fifty templates.
• Schema as a ranking trick. "Structured data isn't required for generative AI search." It's a clarity tool; use it for that. What it's actually for is here.
• Vendors with "insider" AI metrics. "No third-party tool has access to our internal ranking or AI systems." Any dashboard claiming otherwise is estimating, which is fine as long as it says so.

The honest version of this whole field is that there's no trick. The engines name the businesses they can identify, verify and quote. Most of your competitors haven't done the identity work, haven't built the corroboration on purpose, and haven't written a single quotable sentence. The playbook above is a month of unglamorous work and it's where the shortlist comes from.

Questions I get about this