How to Get Your Business Cited by ChatGPT, Perplexity and AI Overviews

How do AI assistants decide what to cite?

Different systems work differently, but they converge on a similar shortlist. A model answering a live question retrieves candidate sources from a search index, ranks them for relevance and reliability, then generates an answer from the ones it can extract clean, specific statements out of.

That gives you three levers, in order of how much control you have:

1. Retrievability — can the system find and read your page at all?
2. Extractability — can it lift a clean, self-contained statement out of it?
3. Trust — does everything else it knows about you agree with what the page says?

Most businesses focus on the third and neglect the first two. A trustworthy business with a JavaScript-rendered site and no clear answers on the page is invisible to the process before trust is ever assessed.

What makes a page extractable?

Extractable content shares a recognisable shape. Each section can be lifted out and still make sense with nothing around it.

• A question-style heading that matches how a person would actually ask it.
• A direct answer in the first 40–100 words underneath, with no preamble and no 'in this article we'll explore'.
• Then the supporting detail — a list, a table, an example.
• No back-references. 'As mentioned above' breaks the section the moment it's extracted.
• Specific nouns and numbers rather than pronouns and generalities.

Which platforms feed AI answers about local businesses?

For a question like "best commercial electrician in Christchurch", the answer is assembled from structured business data across several ecosystems — not from your website alone.

The rule across all of them is consistency. Same legal name, same address format, same phone number, same service descriptions. Where these disagree, the systems have to pick a winner, and the safest choice for them is often to cite a competitor whose data agrees with itself.

We covered the Microsoft side of this in more detail in why Bing can no longer be ignored for local SEO.

What content earns citations that competitors can't copy?

Generic content has no citation value, because a model has fifty equivalent versions to choose from. Four content types reliably earn citations for NZ and Australian businesses:

• Local specifics. How something works under New Zealand law, in the New Zealand market, or at New Zealand prices. There are far fewer credible sources for this than for the equivalent US question.
• Original observations. What you see across the clients you work with, clearly identified as your own observation rather than dressed up as research.
• Structured comparisons. Option A versus option B with criteria and trade-offs, in a table. This is the single most extractable format there is.
• Genuinely transparent cost and process information. Most competitors won't publish it, which is precisely why the pages that do get cited.

One warning that matters more in this context than any other: never invent a statistic, a case study, a client result or a quote to fill a gap. Fabricated specifics are the fastest way to become a source that gets contradicted by other sources — and a contradicted source stops being cited. If you can't verify a number, describe the mechanism instead and say plainly that you don't have the number.

What technical setup helps AI systems find you?

1. Server-rendered HTML. If the content only exists after client-side JavaScript runs, treat it as unreliable for extraction.
2. A clean robots.txt that doesn't accidentally block AI crawlers you actually want, and an accurate XML sitemap.
3. Structured data — Organization, Service, FAQPage, BlogPosting, BreadcrumbList — generated from the visible page content, never written separately.
4. Stable canonical URLs. Changing slugs without redirects destroys accumulated citation value.
5. Fast pages. Retrieval systems have timeouts like anyone else.
6. Descriptive internal links, so the relationships between your pages are legible.
7. An llms.txt file summarising what your organisation does and pointing to your key pages — cheap to publish, and it must agree with your website exactly.

How do you measure whether it's working?

There is no rank tracker for AI citations yet, so build a lightweight baseline you can actually maintain:

1. Write down twenty prompts a real customer might use — including the ones with your city in them, and the ones that name competitors.
2. Run them across ChatGPT, Perplexity, Gemini and Google AI Overviews. Record whether you're named, what's said, and who is named instead.
3. Re-run the same twenty prompts every quarter, unchanged. Changing the prompts destroys the comparison.
4. In GA4, watch referrals from AI domains — the volumes are small, but the intent is often unusually high.
5. In Google Search Console, track branded impressions. A rise in people searching your name is often the first visible sign an assistant is recommending you.

Expect answers to vary between runs — these systems are not deterministic, and a single absence isn't a trend. What you're looking for is the direction across twenty prompts over several quarters, not a result on any one.

What should you do this month?

• Run the twenty-prompt baseline. It takes an afternoon and it's the only way to know where you stand.
• Claim and complete Bing Places if you haven't — it's the biggest quick win for most NZ businesses.
• Audit your business details across your site, Google Business Profile, Bing Places and your three biggest directory listings. Fix every discrepancy.
• Add a genuine 40–100 word direct answer to your five most commercially important pages.
• Publish one piece of genuinely local content only your business could write.