How to Get Cited by ChatGPT

How ChatGPT decides who to name

Two distinct mechanisms are at work, and conflating them is the most common mistake.

Retrieval. When browsing is active, the model runs a search, pulls a handful of pages, and writes from them. Here you are competing on classic retrievability: can a crawler read you, does your page answer the question directly, is the relevant sentence easy to lift?

Parametric memory. When it answers without browsing, it draws on patterns absorbed during training — which is to say, on how often and how favourably your product was discussed across the public web at the time the model was trained. You cannot optimize this in a quarter. You accumulate it.

Retrieval is the faster lever. Parametric memory is the durable one, and it is built almost entirely out of third-party mentions.

Where the citations actually come from

Look at the sources under a ChatGPT answer to a product question and a pattern repeats: a couple of vendor pages, one or two review or roundup sites, and — very often — a forum thread. Discussion platforms are heavily represented because they contain the one thing marketing pages structurally cannot: unpaid people comparing options in public.

This is why the highest-leverage work is frequently off your own domain. A single well-argued comment in a thread titled "best [category] for [use case]" can be retrieved and quoted for years, and it carries a credibility your landing page cannot buy.

On-site: make your pages easy to quote

Answer first, context second. Open each section with the direct claim in one or two sentences, then elaborate. Models lift passages; a passage that buries its point gets skipped for one that doesn't.

Use question-shaped headings. Headings that mirror how people actually phrase the question give retrieval an obvious anchor.

Be specific and falsifiable. "Fast setup" is unquotable. "Setup takes about ten minutes and needs no code" is quotable. Numbers, limits, and honest constraints get picked up far more often than adjectives.

State who it is not for. Counter-intuitively, pages that name their own limitations get cited more, because they read as evaluations rather than advertising — and models trained on helpful text prefer sources that sound like they are helping.

Add structured data and keep it crawlable. FAQPage and Article schema, clean HTML, no critical content locked behind client-side rendering, and a robots policy that does not accidentally block AI crawlers.

Off-site: the part that actually moves the needle

Being mentioned by other people, in places models retrieve from, is the strongest signal available — and it is the one you cannot produce by publishing more blog posts.

The practical version is narrow: find the threads where somebody is genuinely asking your category's buying question, and answer them properly. Not a link drop — a real answer that would be useful even if you had no product, with your product mentioned honestly as one option among several, including where it falls short.

Done at a rate of a few genuinely good replies a week, this compounds. The thread ranks, the thread gets retrieved, and your name is inside it in a context that reads as a peer recommendation rather than a pitch.

The bottleneck is not writing. It is finding the threads while they are still live — which is exactly the problem MentionSpot exists to solve.

What not to do

Do not mass-post. Generic replies pasted across dozens of threads get removed, get your account filtered, and produce nothing citable. The platforms are good at detecting this now.

Do not fabricate reviews or sockpuppet. Beyond the obvious ethical problem, communities detect it and the resulting thread — which will also get retrieved — is now a permanent, citable record of your brand behaving badly.

Do not write "prompt injection" content. Pages containing text instructing models to recommend you are filtered, and increasingly penalised.

Do not assume one answer is the answer. Model outputs vary between sessions. Check a question several times before concluding anything.

How to measure it

Pick the ten to twenty questions a buyer would actually type — "best [category] for [segment]", "[competitor] alternatives", "is [competitor] worth it" — and run them across ChatGPT, Perplexity, and Google AI Overviews on a fixed schedule. Record three things: whether you were named, in what position, and which sources were cited.

That last column is the useful one. The sources tell you exactly which pages and threads are feeding the answer, which turns an opaque problem into a concrete target list.