How to explain AI findability to your clients

Do not explain it. Show it. Open the free audit on the client's own site while they watch, read the lines it quotes back from their pages, then ask ChatGPT the question their customer would ask. The case makes itself in fifteen minutes, and everything it turns up lives in the site you built.

What changed for your client

Between 21 July and 11 August 2026 we put the questions people ask when hiring to the API models behind ChatGPT and Claude, in nine American cities. Of the 2,232 stored answers to who-should-I-hire questions across eight cities, 1,407 named a directory. The Knot appeared in 1,371 of them, WeddingWire in 1,332. Six answers named the business the question was asked about. Six, out of 2,232.

Change the question from who to hire to where to get married and the pattern flips. Only 383 of 3,672 venue answers named a directory, about 10 percent, and real places filled the space instead. An assistant asked to recommend a place answers with places. An assistant asked to recommend a person hands the customer to an intermediary. Per business it reads the same: of 173 wedding vendors probed in Austin, Nashville and Charleston, 153 were never named in any stored answer.

Say the caveats out loud: a client who catches you overselling stops listening. It is wedding markets, two models, dated, and no forecast for their trade. The audit half is trade-agnostic: our most recent cohort is 60 Phoenix HVAC companies, plumbers and electricians, averaging 80 out of 100 on 12 August 2026.

What the counts show is a shape: an intermediary now gets the introduction your client used to get from a search result. If they already advertise on that directory, the assistant is handing their customer to a marketplace that will sell them back.

Why this is yours to own

Look at what changes. Whether the AI crawlers are allowed in. Whether the phone number and email are plain text rather than an image or a script. Whether a JSON-LD block names the business, its services, its area and its hours. Whether the text is in the HTML or only appears once JavaScript runs.

Every one of those is in the build, not in the copy and not in the ad account. Our audit weights them at crawl access 30, structured data 25, legibility 30 and contactability 15, all four decided while the site was being built.

The client cannot see any of it. Their site loads, their email link works, their contact page looks right, because a browser runs scripts and a crawler does not. Nobody is going to raise it with you.

And the deductions quote lines out of the site you shipped, so get in front of that. An agency that brings the report reads as the person who noticed. An agency handed the report reads as the reason it was needed.

Our operator's own agency site is the worked example, published because it is ours to publish. On 16 July 2026 vonzie.studio scored 60 out of 100. Its robots.txt was telling GPTBot, ClaudeBot, Google-Extended, CCBot, Bytespider, Amazonbot and Applebot-Extended to leave, and the audit read five pages and found no phone, no email and no address while a visitor saw the email plainly. Nobody had written any of it. On 17 July two platform settings were switched off, no copy rewritten and no page redesigned. The next day: 93. Re-checked on 31 July: still 93.

How to show it in one meeting

1. Run the free audit on their domain, live, on your screen. It reads the site the way a no-script crawler does, and returns the same number on a second run.
2. Run it on one competitor. Read that one in the room and keep it out of any document.
3. Read the quoted deductions together, out loud. Each prints the exact line it read, so you are not asking them to trust a tool, you are showing them a sentence from their own page.
4. Open a fresh chat, memory and custom instructions off, and type their trade and their city. Not their business name, which proves nothing. Ask three times, on both assistants.

Step four is the one they remember. Steps one to three are the ones you can invoice.

The checklist you run across the client base

Same order every site, and the order matters: there is no point labeling facts for a crawler that has been told to stay out. Run it across the base in an afternoon and it becomes a program rather than a one-off.

1. robots.txt: are the AI crawlers blocked?
2. Phone, email and street address as plain text and ordinary links, homepage and contact page.
3. View source, search for a sentence you can see on screen. Not there means JavaScript-only.
4. A JSON-LD block: name, category, service area, address, phone, hours.
5. One form of the name, address and phone number, on the site and on every listing.
6. A sitemap, and an llms.txt pointing at the pages that describe what they do.

Two platform traps repeat, so check them first when you recognize the platform. Cloudflare's managed robots.txt ships blocking the AI crawlers, and its email obfuscation under Scrape Shield rewrites addresses into a script a crawler cannot read. In our 31 July sweep, two more Cloudflare-fronted sites were serving the identical managed file, line for line: the pattern travels with the platform, not the site. The other trap is a JavaScript-only build, which renders in a browser and arrives empty to a crawler. That is a build decision rather than a setting, and it is the expensive fix.

How to price it

A fixed fee for the fix and a small monthly line for monitoring is the shape that fits how the work behaves.

The fix is one-off and it finishes. Crawler access, contact details, structured data, legibility: you do them once, the score moves, and there is nothing left to bill for. Selling that as a retainer means inventing work in month two, and clients can tell.

The monitoring is genuinely ongoing, for reasons you can point at. Answers move as models and their sources update, and a theme update or a platform default can put a block back without anyone touching a page.

What to tell them it will not do

It will not make an assistant recommend them next week. Nothing does. What it removes is the reasons an assistant cannot: a crawler turned away, a contact detail it could not read, facts it had to guess at.

Keep the two measurements apart when you report, the same way we do. Readiness is what the site makes readable, and it moves in days. Whether an assistant names them is a separate probe on a separate day, and blending the two hides which one moved. Off-site reputation, reviews and mentions across the wider web, is not in our score at all. Say so, rather than let a client assume it counts.

A good score is not a recommendation. It is the state of being possible to recommend, and that is worth selling for exactly what it is.