Key takeaways
- AI answers lean on a predictable set of third-party sources: Reddit, YouTube, reviews, Wikipedia and 'best of' lists. Here is how UK firms earn a place.
- The same source types decide most AI answers: community forums, video, review platforms, reference sites and editorial roundups.
- Reddit, Wikipedia and YouTube top the citation charts, but each AI tool leans on a slightly different mix, so spread your presence.
- You earn these placements the honest way: be genuinely useful, collect real reviews and keep your facts straight. Spam gets filtered out.
- Your own website is rarely what AI quotes. What other people say about you, in the right places, is what gets you named.
When an AI assistant recommends a business, it is usually repeating what a handful of trusted third-party sources already say. Get cited by those sources and you get named. Stay off them and you stay invisible.
Ask ChatGPT, Perplexity or Google’s AI Overviews for the best anything, and the answer does not come from thin air. It is assembled from pages the model retrieved and decided it could trust. Look at what those pages actually are and a pattern appears fast: the same kinds of sites show up again and again, across almost every question and almost every tool.
That pattern is good news for a UK business, because it means the job is not a mystery. You do not need to be everywhere. You need to earn a credible presence on the specific sources AI leans on, in the honest way that holds up when a machine cross-checks you. This guide names those sources, explains why they carry weight, and shows how to earn a place without resorting to spam.
It sits inside our wider guide to answer engine optimisation. Where our piece on UK directories and citations covers the structured listings that confirm you are a real business, this one is about the broader set of sources AI quotes in its actual answer: community discussion, video, reviews, reference sites and “best of” roundups.
Which sources do AI assistants actually pull from?
A short, predictable set: community forums, video, review platforms, reference sites and editorial “best of” lists. The same names dominate study after study, which is exactly why they are worth your attention.
In a Semrush analysis of more than 150,000 AI citations, the most-cited domains were Reddit (40.1%), Wikipedia (26.3%) and YouTube (23.5%), with review and local platforms like Yelp close behind. User-generated content, the material written by ordinary people rather than by brands, makes up most of the top ten. The through-line is simple: AI trusts places where lots of independent people have already weighed in.
For a UK SMB, that reframes the whole task. The pages you polish hardest, your own homepage and service pages, are rarely the ones a model quotes. The sources doing the deciding are the ones you do not own and cannot edit, which is precisely why a machine believes them.
Why do these third-party sources carry so much weight?
Because AI will not stake a recommendation on your word alone. It wants independent corroboration, and these sources provide it at scale. An assistant that names the wrong business looks foolish, so it leans on evidence it can check.
Three qualities make a source citable. It is independent, so it reads as honest rather than promotional. It is consistent with everything else the model finds, so the facts line up. And it is easy to parse, so the machine can lift a clean, quotable line. Community threads, reviews and reference entries tick all three, which is why they outweigh the marketing copy on your own site.
Picture someone asking an assistant for a good accountant in Leicester. The model does not read ten brochures and pick the glossiest. It scans a review profile with recent feedback, a forum thread where locals name names, perhaps a “best accountants in Leicester” roundup, and a reference entry confirming the firm is real. Where those agree, it can recommend with confidence. Where a business appears only on its own website, it has nothing to check against, so it stays quiet.
There is a catch worth naming. Because these sources are open, some people try to game them, and the models have learned to discount obvious manipulation. Fake reviews, planted forum posts and thin listicles built to sell links tend to get filtered, not rewarded. The businesses that win are the ones with a genuine, earned presence that holds up under scrutiny. For the full picture of how AI chooses what to name, our guide to getting recommended by ChatGPT walks through the retrieve, verify and attribute test in detail.
The sources that matter, and how to earn a place on each
Here is the predictable set, what makes each one trustworthy to a machine, and the honest route onto it. You do not need all five firing at once. Pick the two or three where your customers already spend time and do those properly.
| Source type | Why AI trusts it | How to earn a credible presence |
|---|---|---|
| Community forums (Reddit, Quora) | Unscripted peer opinion at scale, hard to fake convincingly, with high engagement | Take part honestly in threads about your field, answer real questions, and let satisfied customers mention you. Never astroturf. |
| Video (YouTube) | Transcripts and descriptions machines can read, and it shows the real thing in action | Publish clear how-to and explainer videos, with accurate titles, full descriptions and captions so the content is readable. |
| Review platforms (Trustpilot, G2, Yelp) | Independent, verified sentiment with the recency and volume a model can weigh | Claim your profiles, collect a steady stream of recent, specific reviews, and reply to all of them, including the awkward ones. |
| Reference sites (Wikipedia, Wikidata) | Structured, corroborated entity facts that feed the knowledge graph directly | Keep an accurate Wikidata entry, which any verifiable business can have. Pursue Wikipedia only if you are genuinely notable. |
| “Best of” roundups (editorial “best X in town” lists) | Pre-filtered, relevance-matched shortlists an AI can lift almost whole | Earn honest inclusion through genuine merit, data or outreach to the writer. Avoid paid link farms, which carry no trust. |
A quick note on fit. A local trade or service business should lead with reviews and the odd local roundup. A B2B or software firm should weight YouTube, G2 and the industry lists its buyers read before they shortlist. Almost everyone benefits from a clean reference entry and a real presence in the communities where their customers ask questions.
How do you earn a presence without spamming?
Slowly, specifically and in public, the same way you would earn a reputation offline. There is no shortcut that survives contact with a model trained to spot manipulation, so put the effort where it compounds.
Be genuinely useful on Reddit and Quora
Show up as a knowledgeable human, not a billboard. Find the subreddits and threads where your customers already ask questions, and answer them properly, with no hard sell. Over time, mentions of your business by you and by others become the kind of signal AI reads as real.
Put real answers on YouTube
Video is climbing fast, and the machines can finally read it. YouTube now appears in 16% of LLM answers against Reddit’s 10%, partly because transcripts make videos easy to parse. Record short, honest how-to and comparison videos, then write full descriptions and captions so nothing useful is locked inside the footage.
Build steady, specific reviews
A trickle of recent, detailed reviews beats a wall of old five-star ratings. Ask every happy customer, across Google and the one or two platforms that carry weight in your sector. Reviews that name the exact job, the town and the outcome give a model quotable language to work with, which counts for far more than a bare star rating.
Get your entity facts straight
Make sure the reference sources describe one consistent business. Confirm your Wikidata and Companies House details are current, and that your name, location and category match wherever they appear. The structured facts here feed the knowledge graph that the rest of your presence hangs off.
Earn honest inclusion in “best of” lists
Be the business a writer can justify including. Pitch the journalists and bloggers who compile “best service in town” roundups with something real: a track record, useful data, a genuine angle. One honest mention on a respected list can outweigh a hundred paid directory slots.
Do the AI tools all pull from the same places?
Mostly, but not identically, so do not bet everything on one source. The big names appear everywhere, yet the mix shifts from tool to tool and month to month.
One analysis of citation patterns found Wikipedia made up nearly 48% of ChatGPT’s top sources while Reddit made up almost 47% of Perplexity’s, with Google’s AI Overviews spreading its weight across Reddit, YouTube and Quora. The mix is also volatile: Semrush tracked ChatGPT citing Reddit in close to 60% of answers one month, then around 10% weeks later after a sourcing change. The lesson is not to chase one platform’s current favourite. It is to build a genuine footprint across the whole trusted set, so a shift in any single source does not wipe you off the map.
Where should a UK business start?
Start with the one or two sources your customers already trust, then widen out. You cannot build everything at once, and you do not need to.
A sensible order for most SMBs:
- Fix your reviews first. Claim your profiles and start collecting recent, specific ones at a steady pace.
- Confirm your entity facts, so Wikidata, Companies House and your listings all agree on the same business.
- Add one honest channel a month: a few YouTube answers, genuine help in the right communities, a pitch to a relevant “best of” writer.
Do that and you are earning a presence on the sources that actually shape AI answers, without a single spammy shortcut.
If you would rather not pick through all of this yourself, it is the work we do. We run the audit that shows exactly where AI cannot see you, then build the review, community, video and reference footprint that gets you cited. It is scoped monthly, with weekly updates and no lock-in, so you can see it working before you commit. Book a free growth review or just get in touch, and we will show you where you stand. We ran the same test on ourselves first and came back invisible, so we know exactly where to look.