Getting cited: the sources that keep coming back

The sources an engine leans on are neither secret nor exotic. That is what makes this a task you can finish rather than a strategy you maintain.

The shape of the list

When an assistant names local businesses, it is usually working from a handful of recurring places: the large horizontal directories, one or two aggregators specific to the trade, community threads where people ask for recommendations, and the business’s own site.

Which ones, exactly, depends on your trade and your country — a home-services aggregator that dominates in one market is unknown in the next. So the list below is a shape to go and verify, not a ranking to trust.

We do not publish a league table of sources, because we have not measured one that would hold across every trade. Your own report names the sources that came back for your prompts, which is the version of this list worth working.

That refusal is worth a sentence more, because the league tables are everywhere. A list of the ten sites that matter most for AI visibility is easy to write and impossible to justify: the sources an engine reaches for shift by category, by country, by city size and by the question being asked. A directory that decides answers for restaurants in one country may never appear in an answer about roofing in another. Anyone publishing a universal ranking has either measured something very narrow and generalised it, or measured nothing.

Why this is finite work

Here is the good news, and it is genuinely good.

The list of sources that describe you is short, it is knowable, and it stops. This is not a content strategy you feed forever or a ranking you defend weekly. It is a few afternoons of records, done once, then checked occasionally. Almost nothing else in marketing has that property, and it is the main reason this work pays better than its reputation suggests.

The catch is that it is boring, and boring work gets postponed in favour of interesting work that matters less.

Working it

1. Find your list. Ask an engine the questions your customers ask, and note what it cites. Do this without naming your business — naming it changes the answer.
2. Claim every listing on it. Unclaimed records are the ones that go stale.
3. Make the description identical. Not similar. Identical. Same name, same service words, same towns. Where records disagree, the engine has to choose, and it will not ask you.
4. Earn one or two mentions. A single thread where a real person names you is worth more than a page of copy you wrote about yourself.

The rest of this page is those four steps at the length they actually need.

Finding your list without fooling yourself

Ask the questions a stranger would ask. The best in your trade in your town, who to call, who is open now, what the job costs. Ask each one in a few different assistants, and write down every source that gets cited — not the businesses named, the sources underneath them.

Two rules keep the exercise honest. Do not put your own name in the question, because naming a business hands the engine the retrieval term and manufactures a mention that would not have happened. And ask each question more than once, since answers vary between runs and one citation list is a single sample rather than a pattern.

What you are looking for is repetition. A source that appears once is noise. A source that appears across several questions and several engines is part of how you are described.

Claiming the records

Unclaimed listings are where wrong facts live longest, for a simple reason: nobody is responsible for them. They were generated from an old data feed, or created by a customer, or scraped from a defunct aggregator, and they will sit there being confidently wrong for years.

Claiming is tedious and mostly free. Expect a verification step — a postcard, a phone call, a code — and expect at least one record you cannot claim because it was created against an address or a phone number you no longer use. Those are worth chasing through the platform’s support flow rather than abandoning, because a stale record you do not control is exactly the thing that turns up in an answer six months from now.

Work in the order the engines gave you. A record that never appears in any answer about you is not urgent, whatever its domain authority.

Identical, not similar

This is the step people skip, and it is the one that decides the outcome.

Your business name should be the same string everywhere — not the legal entity in one place and the trading name in another, not with a town appended in some listings to help with search. The address should be formatted the same way, the phone number should be the same number, and the service words should be the same words. If you call it drain unblocking on your site and drain cleaning on a directory and emergency plumbing on a third, you have given an engine three descriptions to reconcile.

Reconciling is what it will do, and the result is a blurred summary that matches nobody’s search exactly.

The same applies to towns. Decide which places you cover, write them in the same order in every record, and do not quietly expand the list on one platform because its field allows more entries.

A contradiction between your site and your profile is more expensive than either one being wrong alone, because it discredits both sources at once. Where the facts in question are hours or contact details, the specific mechanics of tracking one down are in fixing what AI gets wrong about your hours.

Earning a mention that is not yours

A community thread where a real person recommends you carries weight that your own copy cannot, because it is somebody else asserting it.

You cannot manufacture this and should not try. Fake threads are obvious, they get removed, and being caught is worse than being absent. What works is duller: be findable when somebody asks in a local group, answer questions in your trade without pitching, sponsor the thing your town actually does, and get written up once by a local publication that lists businesses in your category.

One or two of these is enough. This is not a channel to scale.

Why the order matters

Most people do step four first, because it is the interesting one. It is also the one that fails when steps two and three have not happened — a mention pointing at a record that says you closed at five when you close at nine just spreads the wrong fact further.

Get the sources agreeing first. Then go and be memorable.

Set a reminder to repeat the exercise a couple of times a year. Records drift, new aggregators appear, and a platform you claimed will eventually change a field you did not ask it to change. Checking twice a year is enough to catch that before it has been repeated in answers for a season.

Your own site belongs on this list

Everything above is about records other people host. The other half of the citation picture is whether anything on your own domain gets pulled into an answer at all — and if the answer is never, the directories are writing your description without you.

That is a separate job with its own mechanics, covered in making your own site the source engines cite. Do it after the records agree, not before, or your pages will simply be the fifth voice in an argument the majority is winning.

Different sources contribute different things

It helps to stop thinking of citations as votes and start thinking about what each one actually supplies to a sentence.

A directory record supplies fields — name, address, phone, hours, category, area. These are what an engine reaches for when it needs to state something checkable, and they are why a wrong field propagates so far: it is the part designed to be copied.

A review platform supplies adjectives and a rating. This is where the tone of your description comes from, and it is why two businesses with identical listings can be written up quite differently.

A community thread supplies the thing neither of the others can: somebody unaffiliated saying you were good. Engines treat that as evidence of a kind, and it is the only source on the list you cannot fill in yourself.

Your own site supplies specificity — the job, the price, the towns — which is exactly the material the other three are thin on.

The practical consequence is that these are not interchangeable. Fixing four directory records will not improve how warmly you are described, and collecting reviews will not correct an address. Match the work to the gap the report actually shows.

The records you will not be able to fix

Some will not yield, and it is worth knowing which before you lose an afternoon.

Duplicates created years ago against a former address, listings on platforms that have been quietly abandoned by their owners, records assembled by data brokers that feed other directories without a front door of their own. Every one of these can still turn up in an answer.

Two moves are worth making. Report duplicates through whatever mechanism the platform offers, which is often slow but does eventually work. And ignore anything that never appeared in the citations your own questions produced — a wrong record nothing reads is a wrong record that costs you nothing, and treating the whole internet as your responsibility is how this becomes infinite work instead of finite work.

Knowing whether it worked

Two numbers move, and neither moves quickly.

The first is how many distinct sources carry information about you — breadth, which saturates once you have a solid handful and stops paying after that. The second is what share of everything cited about you came from your own domain. How both are scored is worth reading before you judge your own progress, because the first number has a ceiling by design and watching it flatten is not the same as stalling.

Give it weeks rather than days. Records propagate slowly, engines re-read on their own schedule, and the work you did on a Tuesday shows up in an answer whenever the source that carries it is next pulled.