Your Server Cannot Ask ChatGPT a Question

You just watched a plugin grade what it calls 'AI visibility' before your coffee finished cooling, and that speed is the whole tell: whatever it measured wasn't whether ChatGPT recommends you to a stranger, because that answer was never sitting on your server waiting to be read. We built the plugin doing the grading and the platform underneath it that actually asks the engines, so we know exactly where the first one's receipts run out and the second one's begin.

Here's what trusting the wrong number costs: telling yourself you're covered on the one question that decides whether a customer with money in hand finds you this year, when all you actually confirmed is that your schema validates and your title tag runs the right length. Yoast and Rank Math build real tools for that job, and we know exactly where what they check ends and what they can't check begins, because we built the same local checks. Neither has ever sat in the room where a model decides what to say about you, because your server isn't that room.

Two Jobs, One Login Screen

Your title tag lives on your server. So does your schema, your sitemap and your redirect rules, along with every image on the page missing an alt attribute — the plugin reads all of it the moment you hit save, because reading your own database costs nothing and takes no time at all. We read it that fast because there's nothing else to check.

Ask that same plugin whether Gemini names your business when a stranger asks it a buying question, and it has nothing to read. That conversation happened between somebody nobody on staff has met and a model that has never visited your site, and no amount of PHP finds that out by inspecting its own tables. We have to put the actual question to the actual engine and write down what comes back, word for word, not a summary of it. We built it that way on purpose, so a claim about what an engine says never rests on anything but the engine's own words.

What a Server Can Actually Tell You

The local half of this work is real, and we'd be selling it short to call it a consolation prize. A redirect manager that handles 301, 302, 307, 410 and 451 responses, with CSV import, is a technical-SEO tool a site needs whether or not it ever asks an AI engine anything. A 404 monitor with one-click redirect catches lost traffic every week a site stays live, and an internal-link finder that surfaces your own posts mentioning a phrase without linking to the page that answers it is doing work no AI engine could do for you: it's reading your site the way only something with your whole database already open can. We ship all three because a site owner shouldn't have to choose a vendor just to fix a redirect.

None of that needs an outside opinion. It needs a parser and a clock, and your server already owns both, so we ship those checks for free, forever, whether or not you ever connect an account to us. We didn't build the local checks to make the platform look better by comparison. We built them because a site deserves both done right, so you're never stuck choosing between a fast plugin and an honest audit.

What It Cannot

Whether a customer's question to ChatGPT turns up your name isn't a fact about your server; it's a fact about a model, on a given day, answering a stranger who may or may not have search turned on for that particular reply. Getting that fact means asking the engine the question and recording the answer, and that's a platform job, not a plugin job, so it's the job we built the platform to do. A plugin that claims otherwise is grading a proxy and selling it back to you as though it were the real thing. We didn't invent that boundary to make the platform sound necessary; the boundary made the platform necessary, so we built it. We'd have blurred that line already if blurring it sold better, and it doesn't, so we didn't.

The Three Questions Worth Asking Any Tool That Claims This

Where is the number actually computed: on your server, or somewhere that asked the engine? A count built from your own markup is real, but it measures your markup, not what ChatGPT says about you.

Is the answer stored word for word, or only a percentage sitting on top of it? A mention rate you can't trace back to a transcript is a claim you're asked to take on faith, and we don't build it that way. Our free scorecard runs 8 customer questions across ChatGPT and Gemini, three times each, so every lost question opens to one of the stored answers verbatim, the same text the engine actually returned, not our summary of it. The report also names which model answered, per engine, the way this one does, because we'd rather show our work than ask for blind trust.

What happens when the engine call fails? A failed call is a fact about our wallet, not about your visibility, so counting it as a zero would tell a business it's invisible when the real story is a request timed out. We'd rather disclose the count than guess: we exclude failed calls from every number instead, so nothing gets quietly folded into a lower score.

A Plumber in Ohio, For Instance

Say a plumbing company wants to know whether ChatGPT recommends them when someone nearby asks for a plumber. Their title tags are perfect, their schema validates, their sitemap is current, and every local check this plugin runs comes back green. None of that tells them what ChatGPT actually says, because a passing local check measures whether a page is well built, not whether a stranger's question ever surfaced it.

The only honest way to answer that question is to ask it: put the customer's real question to the real engine, more than once, and read what comes back. That's what we built the free scorecard to do: 8 customer questions drawn from the brand's own category and competitors, asked of ChatGPT and Gemini three times each, so the plumber reads the stored answer verbatim instead of trusting a score nobody can check.

Three Runs Is a Signal, Not a Certificate

An AI's answer moves between runs: the same question asked twice can come back differently, because the model is probabilistic and the web it's searching changes under it. A report that hides that motion is prettier than one that shows it, and prettier isn't honest, so ours shows it. And no report, ours included, can trace one specific fix back to one specific sale. AEO is an opaque system, and a vendor who claims otherwise is selling confidence they don't have. We show the spread instead of hiding it, because a range is the honest unit here, not a point estimate.

One more limit worth naming plainly: a readiness audit against the platform runs once every 24 hours, and the free scorecard runs once per site every 30 days. Neither is a live feed, so a number from three weeks ago is still three weeks old no matter how current the dashboard looks. We report the cadence instead of dressing it up as live.

Why We Draw the Line Here, and Where That Leaves You

We sell the audit that sits on the far side of the line this post just drew, so weigh that before weighing anything else here; we're not pretending to be neutral about which half of this work is worth paying for. We'd rather lose a sale to a reader who checks our math than win one from a reader who doesn't, so we show the math. Read what a plugin computes locally against what only an off-site audit can show, ask the three questions above of anything that claims to measure AI visibility, and read the actual transcript before trusting anyone's percentage, ours included.

We built the plugin free because your title tags and your schema deserve a good local tool whether or not you ever pay us another cent, and we built the platform because somebody has to hold the accounts, pay for the calls, and write down what the engines actually say. Get the plugin if you haven't, and start the free scorecard to see what ChatGPT and Gemini already say about you today. Or read what Yoast, Rank Math and AIOSEO can and can't see first, and how a model actually picks its winners once you're covered.