When a customer asks ChatGPT for a plumber, are you the answer?
The common pattern: a business wins its own name and loses the questions a new customer actually asks. See which questions you win and lose — free, two minutes, no account, every answer word for word.
Free check: 7 questions, ~2 minutes, no account. Full report: $749 · one re-check included · 30-day refund · see the sample
| What a customer asks | GPT | Cld | Plx | Gem | Grk |
|---|---|---|---|---|---|
| emergency plumber Portland | ChatGPT: pending | Claude: pending | Perplexity: pending | Gemini: pending | Grok: pending |
| best plumber in Portland OR | ChatGPT: pending | Claude: pending | Perplexity: pending | Gemini: pending | Grok: pending |
| who should I hire for a burst pipe | ChatGPT: pending | Claude: pending | Perplexity: pending | Gemini: pending | Grok: pending |
| how much does a plumber cost | ChatGPT: pending | Claude: pending | Perplexity: pending | Gemini: pending | Grok: pending |
| drain cleaning near me | ChatGPT: pending | Claude: pending | Perplexity: pending | Gemini: pending | Grok: pending |
| Rivera & Sons reviews | ChatGPT: pending | Claude: pending | Perplexity: pending | Gemini: pending | Grok: pending |
Click any cell to read what that platform actually said.
What you're looking at
Each row is a question a customer types. A ✓ means that platform named the business. ✕ means it answered with somebody else. Click any cell to read the actual answer.
One trade, one city, 81 answers: ChatGPT named fewer than half the businesses the platforms surfaced between them. Checking one platform shows you less than half the picture.
The method
Most tools ask once. That isn't a measurement.
Suppose a tool asks ChatGPT for a plumber once, and you don't come up. What do you actually know? The platforms don't answer the same question the same way twice, so one ask tells you what happened that one time — nothing more.
Asked once, a miss could mean anything from “never” to “most of the time.” Asked 15 times, it means something.
Asked once
What a single miss honestly tells you about your citation rate. Almost nothing — and most tools report it as a number with no range at all.
Asked 15 times
The same miss at the depth the report runs — 3 repeats on 5 platforms, then widened, because repeat answers agree far more than independent sampling would predict. Still a range, because it should be.
Some of those ranges are still wide. We show them anyway, because a number without one looks more certain than it is.
Free vs full
The same rows, measured instead of glimpsed
The free check asks 7 questions once. The full report asks 30, 3 times each, on all five platforms — up to 450 samples — which is what turns an anecdote into a rate. These rows are from the worked sample report; the business is invented, the arithmetic is real.
| What a customer asks | Free check — asked once | Full report — 15 asks |
|---|---|---|
| 24 hour plumber near me Portland | ✕ not cited | 0% · 95% 0%–34% named instead: Lamplight Home Services, Nordqvist Plumbing, The Drain Wardens |
| average price for water heater install Portland | ✓ or ✕ — depends on the day | 7% · 95% 1%–42% |
| Rivera and Sons Plumbing Portland | ✓ cited | 93% · 95% 58%–99% |
The middle row is the reason repetition exists: one ask would call it either way and both calls would be wrong. Read the complete sample report.
What we sell
Measure. Fix. Measure again.
Everyone has an AI now. Implementation stopped being the scarce thing — knowing what to tell it is. We measure what five platforms actually say about you, rank what to change, and write every fix so you can hand it straight to your own AI. No agency. No maintenance fee. The knowing, ready to act on.
The division of labor: we do the one thing your AI can't — sample up to 450 answers under a controlled method — and your AI does the one thing we shouldn't: touch your site. Chat can implement; it can't measure.
The report
Thirty questions your customers actually ask, three times each on five platforms — up to 450 answers, every one word for word. Who's named instead of you, and how often. Every fix written as a prompt you hand to your own AI.
Once: re-check included · 30-day refund, no conditions.
Monthly: the full report in month one · re-measured every quarter and after major model releases · new questions and AI-ready fix briefs monthly · cancel anytime.
The guarantee
It names at least one competitor being recommended in your place — by name, in the platforms' own words — or you don't pay. And a 30-day refund with no conditions either way.
Yes — you could subscribe, take the report, and cancel. We priced the first month knowing some people will. Stated plainly, so nobody has to feel clever.
A re-check is included with the one-time report and quarterly for subscribers; $249 per run otherwise, from the bottom of every finished report.
The delivered path
Rather have it done? The Answer Plan — $439/mo, hand-delivered while it's new. We research the questions, write the answer pages, publish and host them, and re-measure every quarter. Your AI can write a page; it won't keep a library researched, published, and measured. A person replies within one business day; nothing is billed until you confirm.
Agencies quote this kind of audit in the low thousands and retainers in the low thousands a month. Monitoring tools charge $29 to several hundred a month to chart a number that, for most local businesses, mostly doesn't move. This is $749, once, with the re-check included.
Agencies & multi-location: volume, white-label — hello@citationchecks.com
Before you trust our numbers
What we measured, and what we borrowed
Measured ourselves
The share of US small businesses that refuse a live-answer AI crawler while serving people normally, from a random 300-domain draw. The figure this category usually quotes is 30%. Ours moved four times under checking, and only ever downward.
Borrowed, with its conflict shown
of business/question/platform combinations were never mentioned at all, in the largest dataset published on this — 100,000+ answers across 100+ brands. We use it because we have found nothing bigger. Its author co-founded and holds equity in a platform that competes with us. Never read it as independent; we don't.
The limits of the method are below, and each is spelled out in the methodology, with our classifier's own error rates beside it.
The boundary
What is in scope, and what isn't
We would rather measure one thing properly than four things loosely, so here is the boundary before you decide.
How many real customers saw you. Nobody outside the AI companies can see that. We measure whether platforms cite you for representative questions, which is a proxy — a good one, but a proxy.
Why a given answer came out that way. The models don't explain themselves, and we don't pretend to reconstruct their reasoning.
What will happen if you change something. We can identify the edits your gaps point to, ordered by how many lost questions each affects. Whether they improved your odds is only visible in a re-check — that part is a measurement, not a promise.
Anything about ads. This measures organic citation only.