How to Tell Whether an AI Answer Is Citing Your Website or a Third-Party Source

Sep 18

Open the AI answer's sources, click each citation, and check whether the link resolves to your domain — but that's only the start. This guide gives business leaders a repeatable verification method, explains why results vary between runs, and introduces a four-outcome framework for turning citation checks into business decisions.

Here's the short answer: open the AI answer's source list, click through each citation, and check whether the link resolves to a page on your domain. If the link goes anywhere else — a review site, a directory, an industry publication, a forum thread — you weren't cited. A third party was, and it may have been speaking on your behalf. If your brand is named but nothing links back to you at all, that's a third possibility. And if you don't appear anywhere in the answer, that's a fourth.

Those four outcomes are not the same finding, and they don't call for the same response. Most guidance on this topic stops at "cited or not cited," which is a bit like a golfer checking only whether the ball is on the grass. On the fairway, in the rough, in the bunker, and out of bounds are all "on the grass" — but you'd play each one very differently.

This article walks through how to run the check properly, why the results move around between runs, what analytics can and cannot tell you, and — most importantly for a founder, CEO, or CMO — how to read the result as a business signal rather than a curiosity. What follows describes how one platform, ChatGPT, documents its own behavior as of this writing; interfaces change, so treat specifics as a snapshot rather than a permanent spec.

First, Get the Vocabulary Straight: Citation, Mention, or Neither

Before you run a single prompt, three distinctions will keep you from misreading what you see:

A citation is a link inside or alongside the AI answer that resolves to an actual page. It's inspectable. You can click it, see the domain, and read the page.
A mention is your brand name appearing in the answer text with nothing to click. The model knows of you, but nothing in that answer routes a reader to your property.
No retrieval at all is a possibility worth understanding: sometimes the model describes a company from its training data rather than from anything it looked up in that moment. There's no source to inspect because none was consulted. You can't always tell from the outside whether you're seeing a mention or trained-in knowledge — but the observable signal is the same either way: no resolvable link.

The distinction that trips up most executives is the first one versus the second. Seeing your company named in a ChatGPT answer feels like a win. It's better than absence, certainly. But an unlinked mention is one of the least durable forms of AI-search presence you can have, because nothing anchors it to a page you control. This is closely related to the broader problem of why AI search engines recommend competitors instead of you, which is worth understanding before you start diagnosing individual answers.

The Manual Check: A Method You Can Run This Afternoon

You don't need software to get a first read. You need a set of prompts, a spreadsheet, and about an hour. Here's the procedure:

Write down the questions your buyers actually ask. Not keywords — questions. Include your brand name in some ("Is [Company] a good choice for X?"), and leave it out of others ("Who are the best providers of X in [market]?"). The unbranded, close-to-purchase questions are where the commercial stakes live. If you're not sure where to start, finding the buyer questions your business isn't answering is a useful first step.
Run each prompt across the platforms that matter — ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews are a sensible starting set for most US businesses.
Open the sources for every answer. Each platform surfaces sources a little differently — a panel, numbered footnotes, inline links — so look wherever that interface puts them rather than expecting one layout. In ChatGPT, OpenAI's own documentation explains that responses using web search may include citations, and selecting a citation opens its source.
Check the domain on every link. Yours, or someone else's? This single step answers the headline question.
Read the cited page and confirm it actually supports what the answer said. This step is almost universally skipped, and it shouldn't be. OpenAI's own documentation notes that search results and citations can be incomplete, outdated, or incorrect, and advises opening the source to check that it supports the answer. If the platform's own vendor tells you to double-check its citations, take the advice.
Log everything. A simple table turns anecdotes into evidence. Ours looks like this:

Date	Platform	Prompt	Outcome (1–4)	Cited URL	Does the cited page support the claim?	Test conditions
—	—	—	—	—	Yes / No / Partially	Logged in? Location? Plan tier?

Those last two columns are where most logging templates fall short. "Does the cited page support the claim" matters because a citation to a page that doesn't actually back the statement is weak evidence of anything. "Test conditions" matters because they genuinely change results — more on that below.

One Trap Worth Knowing About: The Sources Panel Isn't Proof

Here's a nuance that almost nobody flags. In ChatGPT, OpenAI's documentation states that the Sources control displays cited sources and other relevant links. That means your page can appear in the panel without having actually informed the answer. Appearing in a source list and shaping the answer are two different things — which is exactly why step five above, reading the cited page against the claim, belongs in your method. Treat "we showed up in the sources" as a promising signal to verify, not a conclusion. This behavior is documented for ChatGPT specifically; treat it as a reason to verify everywhere, not as a claim about how every platform works.

The Four-Outcome Citation Read: What Each Result Actually Means

Once you've run the check, you'll land in one of four states for each prompt. This is where detection turns into interpretation — and where most of the advice on this topic goes quiet. We call this the Four-Outcome Citation Read, and it's the lens we use with every client before we talk about fixing anything. Here's how each outcome breaks down:

Outcome 1: Your Own Domain Is Cited

What it means: Authority on that question currently sits on your property. The AI system retrieved your page and pointed the reader at it.

What to do: Protect and extend. Check whether the cited page is the one you'd want representing you — if the AI is citing a five-year-old blog post instead of your current service page, that's a signal too. This is your strongest position, but it's an observation at a point in time, not a permanent fixture.

Outcome 2: A Third-Party Page About You Is Cited

What it means: Authority on that question exists — it just lives on someone else's property. A review platform, a directory, a trade publication, or a forum is doing the talking, and the AI surfaced their page instead of yours (or couldn't find yours).

What to do: Read the third-party page carefully, because it's now part of your de facto sales team whether you hired it or not. Then treat the finding as what it usually is: an owned-content gap, not a technical failure. The market has likely published material about your category that currently answers the question more visibly than your site does — a pattern worth confirming with a proper content gap analysis for AI search. That's fixable — and it's a content and presence problem, which is a much better problem to have than invisibility.

Outcome 3: You're Named, Nothing Is Linked

What it means: The model appears to know of you without currently sourcing you. You have awareness without an anchor.

What to do: Treat this as one of the weakest, least durable forms of presence. Nothing in that answer gives a buyer a path to your property, and nothing you control is reinforcing the mention. It's better than absence, but only just — and it's exactly the state that tends to lull leadership teams into a false sense of security. "ChatGPT knows who we are" is not a visibility strategy.

Outcome 4: You're Absent Entirely

What it means: For a question you believe you should own, you don't appear — and, in some cases, a competitor does.

What to do: This is the highest-priority state, and it's the one where competitive comparison matters most. Absence on a definitional question ("what is X?") is usually low-stakes. Absence on a question a buyer asks right before choosing a vendor is a different animal. Which brings us to prioritization.

How to Prioritize: Weight by Commercial Proximity, Not Gap Count

A citation check across twenty prompts and five platforms will hand you a pile of findings. The instinct is to fix everything. The better move is to rank your prompts by how close they sit to a purchase decision, then read the citation outcomes against that ranking.

An Outcome 4 (absence) on "best commercial HVAC service providers in Dallas" generally outweighs an Outcome 2 (third-party citation) on "how often should HVAC systems be serviced" — even though the second finding might feel more fixable. The question isn't how many gaps you have. It's which gaps sit in the path of a buyer with a checkbook.

This is, candidly, how we approach the AI Search Visibility Audit at Discovery Authority: a competitive analysis of observed visibility across Claude, ChatGPT, Perplexity, Gemini, and Google AI Overviews, read against competitors and against the buyer questions that carry commercial weight — and delivered as a prioritized roadmap that addresses high-impact gaps first, rather than a generic visibility score. The findings are point-in-time evidence, not a prediction of future AI behavior or a guarantee of future placement. But a prioritized, competitively framed read is a decision input in a way that a one-off manual check never quite is.

Why Your Results Change Between Runs (and Why That's Not a Bug)

Run the same prompt twice and you may see different sources. Nothing is broken. OpenAI's documentation explains that ChatGPT typically rewrites your question into one or more different, more targeted search queries — and may issue additional follow-up queries after reviewing the initial results. Different rewrites can surface different sources. The retrieval path is not a fixed pipe; it can take a fresh route each time.

Two practical consequences for your measurement discipline:

A single check is a snapshot, not a verdict. A small set of observations across the same prompts, taken over time, tells you more than one run ever can. Don't build a budget decision on a single screenshot — in either direction.
Test conditions are variables. OpenAI's documentation notes that general location from your IP address may be shared with search providers to shape results, which means two colleagues running the identical prompt from different cities are not necessarily running the same test. Log where and how each check was run.

How often should you recheck? There's no research-backed cadence, and anyone who gives you one with confidence is guessing. Our practical judgment: check often enough that you're comparing patterns rather than reacting to single answers, and always around the prompts with the most commercial weight.

What GA4 Referral Traffic Can and Can't Tell You

Analytics is worth checking, and worth keeping in its lane. Referral traffic from AI platforms can tell you that someone arrived at your site from that platform. It does not tell you which prompt they asked, what the answer said, or whether your page was the source that shaped it. Analytics records a visit; citation checking records what the answer actually pointed to. They answer different questions, and traffic can corroborate a citation finding — it doesn't prove one. This is a large part of why we treat an AI visibility audit and brand mention tracking as fundamentally different exercises. Use both, and don't let a healthy referral number substitute for actually inspecting the answers your buyers see.

Can You Automate This?

Yes — automated AI-visibility tracking exists, and it's a reasonable way to scale past manual spot checks once you know what you're looking for. But here's the honest take: collection was never the hard part. The hard part is deciding which prompts are worth tracking, holding the test conditions steady, and interpreting what the pattern means for where you spend money next. A dashboard full of citation data with no prioritization attached is just a prettier version of guesswork.

The Bigger Signal: What a Third-Party-Heavy Profile Tells a Leadership Team

If your checks consistently come back as Outcome 2 — AI answers about your category citing review sites, directories, and publications instead of you — resist the urge to read it as a technical problem or a reason for gloom. Read it as a map.

A third-party-heavy citation profile tends to tell you where your owned material is thin relative to what the market has already published. The authority exists; it's just parked on someone else's property. Closing that gap is a content and presence discipline: consistent, genuinely useful, expert-backed material on the questions that matter, published on ground you control. That's the work our Content Authority System is built for — human-reviewed articles and branded LinkedIn and X content aimed at evidence-backed gaps rather than a generic editorial calendar, run as a system so it doesn't collapse under the resource strain that kills most internal content programs.

The sequence matters, though. Evidence first, content second. Producing content against gaps you haven't verified is just guesswork with better formatting.

FAQ

How is AI visibility measured?

By observation: running a defined set of buyer-relevant prompts across AI platforms, recording whether and how the brand appears (own-domain citation, third-party citation, unlinked mention, or absence), verifying that cited pages actually support the claims made, and repeating the process so you're reading patterns rather than single answers. The competitive layer — who is being cited when you aren't — is what turns raw observations into something a leadership team can act on. Every measurement is a snapshot of observed behavior at a point in time, not a permanent state.

Which AI platforms should we monitor?

For most US businesses, a practical starting set is ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews. The right weighting depends on your buyers — but until you've observed where your buyers' questions surface your competitors, monitor broadly rather than betting on one platform.

Does being mentioned without a link count as AI visibility?

It counts as awareness, not as durable visibility. An unlinked mention gives buyers no path to your property and gives you nothing you control to reinforce it. It's better than absence, and it's a reasonable early signal — but treat it as one of the weakest of the four outcomes, not as a finish line.

My competitor gets cited and I don't. Should I panic?

No — but you should investigate. One answer citing a competitor is noise; the same competitor cited consistently across platforms and across your highest-value buyer questions is a pattern worth attention. Patterns are what deserve budget. The useful response is a structured competitive read, not a reaction to a single screenshot.

Is AI search too new to justify measuring?

The measurement is exactly what resolves that question. If your checks show buyers' questions being answered with competitor and third-party sources, the channel is already influencing how your category is being described. If they don't, you've spent an afternoon confirming your position — cheap insurance either way. The point of measuring first is that you never have to take anyone's word for how much this matters to your business.

Start With Evidence, Not Opinions

The question "is AI citing my website or someone else's?" has a knowable answer, and you now have the method: run buyer-relevant prompts, inspect every source, verify the cited pages, log the conditions, and classify each result into one of the four outcomes above. What you do next depends on which outcomes cluster around your most commercially important questions.

If you'd rather see that read done competitively, across platforms, and prioritized by what actually matters to your pipeline, that's precisely what Discovery Authority's AI Search Visibility Audit is built to deliver — observed evidence and a prioritized roadmap, not a sales deck. For teams also weighing traditional search performance alongside AI visibility, our Search & Paid Media services cover technical SEO and Google Ads management under the same senior-led strategy.

Call Discovery Authority to discuss at 925-963-5767 or click to schedule time.

John Spears