The Local AI Visibility Playbook.

Contents

1. 01 · The shift that already happened
2. 02 · How AI assistants decide who to cite
3. 03 · The four engines, briefly
4. 04 · The Visibility Loop
5. 05 · The fixes that actually move the needle
6. 06 · What to skip (it won't help)
7. 07 · How to know it's working
8. 08 · What to do this week

Twenty years ago, when someone needed a plumber, they opened the phone book. Then for fifteen years, they Googled it and scrolled a list of ten blue links. Today, more and more of them open ChatGPT and ask “who’s the best plumber near me?” — and the AI gives them one answer. Maybe two.

That’s the whole shift. From a list of ten to a single name. It changes everything about how a local business needs to position itself online, and most local businesses haven’t noticed.

The numbers people quote about AI search adoption are still small-ish — somewhere between 10% and 30% of searches depending on who’s measuring and which category. But two things are true: (a) it’s growing fast, and (b) the people using AI search first are disproportionately younger, more affluent, and more decisive — exactly the customers most local service businesses want.

More importantly: the businesses that get cited by AI now build durable defaults. When ChatGPT consistently names you for “best HVAC tech in Marietta,” that becomes self-reinforcing — every conversation that produces your name trains future citations. The compounding works in your favor or in your competitor’s. Being early is structurally valuable.

Modern AI assistants build their answers from two sources: training data (what the model learned during training, essentially a snapshot of the public web at some past date) and retrieval (real-time fetching of pages, listings, and structured data when the user asks something current).

For local recommendations — “best [trade] in [town]” — retrieval matters most. The assistant queries its tools, pulls content from sources it can read, weighs trust signals, and picks one or two names to surface. Here’s the part most agencies get backwards: the strongest signals are not on your website. They’re what the rest of the web says about you.

The signals that matter, in order of evidence

• Third-party mentions — the big one. The largest study to date (Ahrefs, 75,000 brands) found that being mentioned by name in credible publications was the strongest signal measured for AI visibility — roughly three times stronger than backlink count. A separate analysis of 25 million AI citations (Muck Rack) found that most citations point at earned, third-party sources — not the business’s own site. AI engines recommend who the web vouches for.
• Review-platform presence. Across 129,000 domains studied (SE Ranking), businesses with profiles on major review platforms were cited by ChatGPT at roughly three times the rate of businesses without them. For a local trade, this is the most buildable signal on the list.
• Entity consistency. Same name, address, phone, and category everywhere you appear — Google Business, Apple, Bing, BBB, trade directories. When the details disagree, the AI can’t confidently connect your reviews and mentions to one business, and the ambiguity costs you the citation.
• Content that’s easy to quote. The one peer-reviewed study on this (Princeton/Georgia Tech, KDD 2024) measured what on-page changes lift AI visibility: adding quotations (~+41%), concrete statistics (~+40%), and cited sources (~+30%). Real effects — a page with direct answers and real numbers beats a wall of marketing copy. But note the tier: these operate on top of the corroboration layer, not instead of it.
• Reachability. None of the above matters if the engine can’t read your site. Content that only renders in JavaScript, AI crawlers blocked by default (Cloudflare does this out of the box), or a site missing from Bing’s index — which ChatGPT leans on — are all documented ways to be invisible while doing everything else right.

What about schema markup and llms.txt — the things most AEO agencies lead with? Google has publicly said its AI features ignore llms.txt and require no special schema. We treat clean structured data as housekeeping (it removes ambiguity, cheaply), not as a citation lever. Section 06 has the full skip list.

None of these signals are secret or magical. They’re all things a local business can build. The reason most don’t is that the agencies they hire are still selling 2018 SEO — or 2025 schema snake oil — and can’t deliver the off-site work that actually moves citations.

There are dozens of AI assistants, but four matter for local recommendations today:

• ChatGPT (OpenAI) — the largest by usage. Uses Bing for real-time browsing, plus its own training data. Tends to be conservative about local recommendations, favoring businesses with strong directory presence and review surface.
• Claude (Anthropic) — has web search and tends to cite sources prominently. Weights authoritative sources (industry press, professional associations, major directories) heavily. Particularly important for legal and medical categories.
• Gemini (Google)— the engine behind Google’s AI Overviews. Has direct access to Google’s entire search index plus Google Business Profile data. If your GBP is strong, you’re already ahead with Gemini.
• Perplexity — built around real-time web search with visible citations. Smaller user base but disproportionately power users. Citations are prominent and clickable, so showing up on Perplexity drives direct traffic faster than the others.

The good news: the work that gets you cited by one mostly gets you cited by all four. The signals overlap. You don’t need four different strategies — you need one strategy executed well.

This is the five-step process we run on every client engagement. We named it “The Visibility Loop” because step 5 feeds back into step 1, and the wins compound month over month. You can run this yourself if you have the time and the technical comfort. We do it for clients who don’t.

Step 01 · Audit

Score where you stand. The mechanical version: open ChatGPT, Claude, Gemini, and Perplexity. Run twenty queries a customer might ask — variations of “best [your trade] in [your town],” “[problem] near me,” “who does [specific service] in [neighborhood].” Note which ones name you, which name competitors, which name nobody. Then run them again on another day — AI answers are volatile, and a single pass is noise, not a baseline. Two or three passes gives you a real starting number.

Most local businesses score zero. That’s the baseline you’re working from. Don’t skip writing it down — you need a before number to know you’ve moved.

Step 02 · Map

Find the queries that matter — not the ones a keyword tool surfaces, but the ones your customers actually ask AI. The two sources we use:

• The People Also Ask box on Google for your trade and town. Those are real questions real people ask.
• Your sales calls. The first three questions every prospect asks you on the phone — those are also the first three things they tried to ask AI before calling.

Map each query to the page on your site that should be the answer. If no page exists, that’s a content gap. Write it down.

Step 03 · Fix

This is the work most agencies skip — because the highest-impact half of it happens off your website, and that half is harder to deliver than a schema plugin. The order of operations, by evidence:

• Review footprint first. Get present and active on the review platforms AI engines read — Google, Yelp, BBB, and the ones specific to your trade. Ask every completed job for a review, and respond to all of them. This is the strongest signal a local business can directly build.
• Earned local mentions. Local news, the chamber of commerce, trade associations, community sponsorships, your manufacturers’ dealer directories. Real third-party mentions are the top-ranked citation signal in every major study — and one mention with editorial weight beats ten directory listings.
• Listings consistency. NAP (name/address/phone) match across Google Business Profile, Apple Business Connect, Bing Places, BBB, and your industry directories. Complete GBP — categories, services, photos, hours. Inconsistencies create exactly the ambiguity that costs citations.
• Rewrite service pages so they’re quotable. Each service page should answer 4–6 of the questions from step 2 — directly, near the top, with real numbers (years in business, response times, price ranges you’re comfortable naming) and sources where relevant. That’s the on-page pattern the peer-reviewed research actually validates.
• Technical pass. Confirm AI crawlers aren’t blocked (check your CDN defaults — each vendor runs separate bots for search and training), make sure important content is real HTML rather than JavaScript-rendered, and register with Bing Webmaster Tools — ChatGPT’s retrieval leans on Bing’s index. Tidy your structured data while you’re in there. Necessary, not sufficient.

Step 04 · Monitor

Re-run your audit queries weekly. Track which engines name you, for which queries, and how that’s trending. We do this in a spreadsheet for clients and email a one-page summary monthly. No dashboard. The metric is binary per query: cited or not cited. Trends emerge.

Important: AI citations are genuinely volatile. The same question gets different answers across days, model versions, and users. A single negative result means nothing; so does a single positive one. The only honest measurement is repeated sampling — the same queries, run multiple times, across weeks. Anyone showing you a one-screenshot “before and after” is showing you noise.

Step 05 · Iterate

AI search shifts. New models ship every few months. Retrieval sources change. Your monitoring from step 4 tells you what’s working — double down. What isn’t — diagnose and adjust.

The compounding is the point. A business that does this work consistently for a year is essentially uncatchable by a competitor starting from zero. AI assistants tend to keep citing whoever they’ve cited before, because that’s what their training data and retrieval bias toward.

If you do nothing else from this playbook, do these five things, in this order. They’re ranked by the strength of the evidence behind them — which means the off-site work comes first, because that’s where the citations actually come from.

1. Build a real review base on the platforms AI reads

Google, Yelp, BBB, and whichever platforms matter in your trade (Angi for home services, RealSelf for aesthetics, Avvo for legal). Businesses with an active presence on major review platforms get cited by ChatGPT at roughly three times the rate of those without. Build the habit: every completed job gets a review ask, every review gets a response within 48 hours. This costs nothing but discipline, and nothing else on this list outweighs it.

2. Get mentioned by sources that aren’t you

Local press, the chamber of commerce, trade associations, your manufacturers’ dealer directories, community sponsorships with a real write-up. Across every major study, earned third-party mentions are the strongest AI visibility signal measured — most AI citations point at earned coverage, not the business’s own site. One real mention with editorial weight beats ten directory listings. And it has to be real: bought or faked mentions are both detectable and risky.

3. Make your Google Business Profile complete and active

Fill out every field. Add photos monthly. Respond to every review (good and bad) within 48 hours. Post weekly. Your GBP feeds Google’s AI directly and anchors the listing consistency the other engines check you against — and while you’re at it, make sure Apple, Bing, and BBB show the same name, address, and phone.

4. Rewrite your service pages so they’re easy to quote

The one peer-reviewed study on on-page changes (Princeton and Georgia Tech, 2024) found the pages AI engines surface more after adding quotations (~+41%), concrete statistics (~+40%), and cited sources (~+30%). In practice: answer the top customer questions directly near the top of each service page, use real numbers — years in business, jobs completed, response times — and write like a person explaining, not a brochure. Same rule for service-area pages: genuine local detail or don’t build them.

5. Confirm AI can actually reach your site

The unglamorous one that silently zeroes out everything above. Check that AI crawlers aren’t blocked — Cloudflare blocks them by default, and each vendor runs separate bots for search and training. Make sure your important content is real HTML, not JavaScript-rendered. Register with Bing Webmaster Tools, because ChatGPT’s retrieval leans on Bing’s index. Tidy your structured data while you’re in there — it’s reasonable housekeeping, just not the lever the schema-plugin industry says it is.

A short list of things you’ll see other agencies and self-help blogs recommend. Some of it is recycled 2015 SEO advice; some of it is newer AEO folklore that the platforms themselves have now publicly debunked. Either way: skip it.

• llms.txt. A proposed file that’s supposed to tell AI systems what your site is about. Google has said publicly that its AI features ignore it, and no engine has committed to rewarding it. It’s harmless — but if an agency leads their pitch with it, that tells you what else they’re selling.
• Schema markup as the secret weapon. Google’s own guidance says no special structured data is needed for AI features; Bing says at most it may help interpretation, with no guarantees. Keep your schema clean as housekeeping — it removes ambiguity about who you are — but nobody should be charging you for markup as the thing that gets you cited.
• Rewriting your content “for AI”. Chunking your pages into robot-shaped fragments, keyword-stuffing “near me” phrases, or restructuring prose because a blog said LLMs prefer it. Google names this in its own mythbusting. Write clear, direct, useful pages for humans — that’s also what gets quoted.
• Buying links or mentions. Bought backlinks and manufactured “earned media” are both detectable patterns, and the corroboration layer only works because it’s hard to fake. You can hurt your own visibility this way. The real version — reviews you earned, coverage you deserved — is the entire point.
• Generating thin content with AI. The irony — using ChatGPT to write five hundred service pages so you can rank in ChatGPT. Thin templated pages earn no reviews, no mentions, and no quotes; doorway pages with the city name swapped have been penalized since the 2010s. Use AI to draft if you like, but a human needs to edit and add real local detail.
• Promises of overnight results. Anyone telling you they can get you cited by ChatGPT “in a week” is selling you something. Realistic timeline: 30–90 days for the first citations to start appearing, 3–6 months for it to become a meaningful share of inbound. And because citations are volatile, any before/after built on a single screenshot is noise, not proof.
• Ignoring Google. Some AEO content frames Google as dead. It isn’t. Google is still where the majority of search happens, and the corroboration work that wins AI citations — reviews, mentions, consistent listings — is the same work Google’s local results have always rewarded. Optimize for both. The work overlaps almost entirely.

Two metrics matter. Everything else is noise.

Leading metric — citation rate by query

Pick your top 20 customer queries. Run them weekly across all four AI assistants — and run each one more than once, because single answers are noisy. What you’re tracking is the rate: out of all samples this month, how often were you named? That rate climbing means the work is working. Still zero after 90 days means something’s broken — usually a reachability problem (blocked crawlers, unreadable pages) or a corroboration gap (thin reviews, no third-party mentions).

Lagging metric — booked jobs from inbound

Track call volume and form submissions to your tracked phone number and forms. AI search inbound is hard to attribute cleanly (no UTM, no referrer in many cases) — what you’re looking for is rising inbound that doesn’t correlate with paid spend or new GBP posts. That residual lift is your AI visibility paying off.

What to ignore

• Impressions and CTR from Search Console. AI search traffic mostly doesn’t show up there. Don’t use Search Console as your AEO scorecard.
• Domain Rating (DR), Domain Authority (DA), other agency vanity metrics. They’re proprietary scores from SEO tool vendors and AI assistants don’t use them.
• Anyone claiming a positional rank on a single AI engine — “you’re #2 in ChatGPT for plumber queries.” That’s a meaningless claim because AI assistants don’t return ordered lists the way Google does, and their answers change between runs. A single query is binary: cited or not. (A composite visibility scoreacross many queries and repeated samples — citation rate, review footprint, mention count, listing consistency — is a different thing and a useful one. We use one. Don’t confuse the two.)

If you read this far, here’s the smallest set of steps that produces a real improvement, in roughly the time you have on a Friday afternoon.

• Open ChatGPT, Claude, Gemini, and Perplexity. Run five queries a customer might ask about your business — twice each, on different days, because answers vary. Write down who they name.
• If you’re not named in any of them, that’s your starting point. If you are, note which engines and which queries — that’s your foundation to build on.
• Text your last five happy customers and ask for a Google review. A steady review cadence is the cheapest, most proven corroboration there is — start the habit today.
• Open your Google Business Profile. Fill in every field that isn’t filled. Add five recent photos. Respond to every unanswered review.
• Pick your most important service page. Add five real customer questions with direct answers — include a real number or two (years in business, jobs completed, response time). Skip the schema-plugin hunt; clarity is what gets quoted.
• Re-run your queries in 30 days — again, more than once — and compare the rate, not a single answer.

That’s the whole foundation. The deeper work — the review cadence, the earned local mentions, the listing consistency across twelve directories, the crawler and index checks, the monitoring cadence, the iteration loop — is what we do for clients who want it done right and don’t want to manage it themselves.

Want this done for you?

We’ll audit your business this week, and fix it the next.

The free audit covers everything in this playbook — your visibility across all four AI engines, your review and mention footprint, your listings hygiene, your competitive position. Then we tell you what we’d fix first.