The one-page version
For a niche B2B firm that gets decent organic traffic but never appears when a buyer asks an assistant for recommendations.
- 1Choose 20 buyer questions you intend to own. Not 200. Twenty you can genuinely answer better than anyone.
- 2Build one asset per question that answers it in the first 100 words and holds evidence a model can attribute.
- 3Manufacture first-party evidence: your own benchmarks, your own campaign numbers, your own named method.
- 4Get mentioned off-site where models already read: directories, comparison pages, podcasts, forums, trade press.
- 5Track citation rate per question monthly, then track whether any of it turns into enquiries. Only the second one pays.
The short answer
A B2B AI search citation strategy is a deliberate plan to become one of the handful of sources an assistant reaches for when a buyer asks a question in your category. It has five parts: the questions you choose to own, the assets you build to answer them, the first-party evidence that makes those assets worth quoting, the off-site mentions that corroborate you, and the measurement loop that tells you whether any of it moved.
The important shift is from volume to selection. Traditional SEO rewarded publishing more. Citation strategy rewards being the most quotable answer to a narrow set of questions. For a niche B2B firm that is good news, because narrow is where you already live.
What a citation strategy actually is
When someone asks ChatGPT or Perplexity "who should I use for LinkedIn advertising in the UK", the model does not scan the whole web. It runs a retrieval step, pulls a small set of candidate sources, and then composes an answer from the passages it finds most useful. Usually that is three to eight sources. That is the shortlist you are trying to get onto.
Two things get you there. Retrievability, which is whether your page can be found and read at all. And quotability, which is whether any passage on it is specific enough to be worth attributing. Most B2B sites fail on the second. They are perfectly crawlable pages full of claims that are true of every competitor, so there is nothing to quote and no reason to name you.
"You do not get cited for being on the internet. You get cited for being the only place a specific sentence exists."
That is the whole strategy in one line. Everything below is how you engineer sentences that only exist on your site, and then make sure models can find them.
Pick the questions you intend to own
Start with a list of 20 questions, not a keyword export. Questions are how buyers talk to assistants and they carry intent that keywords strip out. "LinkedIn ads agency" tells you nothing. "Is LinkedIn advertising worth it for a 15 person engineering consultancy" tells you exactly who is asking and what they are worried about.
"B2B marketing" · "lead generation" · "digital advertising agency"
Enormous, generic, and already answered by publishers with a decade of authority.
"What should a UK manufacturer expect to pay per lead on LinkedIn?" · "Why do my LinkedIn ads show as direct traffic in GA4?"
Specific, answerable with real data, and thin on genuinely good existing sources.
- Take questions from sales calls first. The ones you answer on every discovery call are the ones your buyers are also typing into assistants.
- Score each question on two axes. Can we answer it better than anyone with data we actually have? And would someone asking it be a plausible client?
- Drop anything you cannot answer with evidence. If your answer is an opinion anyone could hold, the model has no reason to prefer you.
- Keep the list to 20 for the first quarter. Depth beats breadth here. Twenty answered properly outperforms two hundred answered thinly.
Build the citable asset for each one
Each chosen question needs an asset that answers it. Sometimes that is a full article, sometimes a section of a service page, sometimes a small tool or calculator. The format matters less than the shape: direct answer first, evidence second, qualification third.
Run every new page through this before it ships.
The exact buyer question appears as an H1 or H2
The answer lands within the first 100 words
At least one number, range or timeframe that is ours
One passage that could be lifted whole and still make sense
A comparison or table where a decision is involved
A named author and a last-reviewed date
Internal links to the two or three related questions
No claim on the page we could not defend on a call
If a page fails three or more of these, it is a brochure page, not a citation asset. Rewrite it or leave it out of the strategy.
Want more enquiries out of the traffic you already have?
Book a 30 minute call. We will look at where your leads come from now and where the gap is between traffic and pipeline.
Manufacture first-party evidence
This is the pillar most firms skip, and it is the one that decides whether the strategy works. Models cite sources that contain information they cannot get elsewhere. If your page restates what three other pages already said, being clearer is not enough. You need something original on it.
The good news is that any operating B2B firm is sitting on original data and does not realise it. You have campaign results, cost benchmarks, win rates, sales cycle lengths, objection patterns and failure modes. Turned into a stated figure with a stated method, each of those becomes a sentence that only exists on your site.
- Publish your own benchmarks. "Across the campaigns we ran for UK B2B firms last year, cost per landing page click sat between £0.45 and £1.80." State the sample and the period.
- Publish your method by name. A named process is quotable in a way an unnamed process is not. Describe the steps, not just the label.
- Publish worked examples. Situation, action, result, with figures. Anonymise the client if you have to; keep the numbers real.
- Publish what did not work. Failure data is rare, credible and heavily quoted. It also makes everything else on the page more believable.
Get mentioned where models already read
Assistants do not only read your site. When someone asks for recommendations, the answer is often assembled from third-party sources: comparison articles, directories, forum threads, trade publications, podcast transcripts. If your firm is absent from all of them, you are unlikely to appear in a recommendation answer no matter how good your own pages are.
- Be on the lists. Category round-ups and directories in your niche are frequently retrieved. Get listed accurately, with the specialism spelled out.
- Answer questions in public. Industry forums, LinkedIn threads and Q&A sites are readable, indexed and often surprisingly influential in retrieval.
- Do podcasts and get the transcript published. A transcript is a text asset with your name attached to a specific claim. That is exactly what a model can use.
- Get quoted in trade press. One quote in a niche publication your buyers read is worth more than ten generic guest posts.
- Keep the description consistent. If you are described five different ways across five sources, no clear association forms. Same specialism, same wording.
"Your own site says you are good at this. Everywhere else saying it is what makes the model believe it."
Measure citations, then measure leads
There is no Search Console for AI search, so measurement is manual and that is fine. The discipline is to test the same questions in the same way on the same schedule, and to keep the log.
One row per question, per engine, per month. A spreadsheet is enough.
Question · exactly as a buyer would phrase it
Engine · ChatGPT / Perplexity / Google AI Overviews
Date tested
Cited? · yes / no
Sources named · which competitors appeared
Which page of ours was pulled
Notes · what the answer got wrong about us
Answers vary between runs and accounts, so treat the log as a trend, not a scoreboard. Movement over three months is the signal.
Then connect it to the thing that pays. Watch direct and branded organic traffic, and add a "how did you hear about us" field to your enquiry form. AI referrals frequently land as direct traffic with no referrer, so the self-reported answer is often the only clean signal you will get.
The question map template
Before any writing happens, build the map. It forces you to decide what you are trying to own and stops the content plan drifting into whatever is easiest to write.
One row per question. Fill it in before briefing a single page.
Question (buyer's exact phrasing)
Buyer stage · problem aware / solution aware / vendor selection
Can we answer with first-party data? · yes / partly / no
Evidence we hold · the actual numbers or examples
Asset type · article / service page section / tool
Existing page or new build
Competitors currently cited
Priority · 1-3
Anything scoring 'no' on first-party data goes to the bottom. You can still write it later, but it will not win citations on its own.
The 90-day rollout
Three months is roughly the point at which you can tell whether the strategy is working. Less than that and you are reading noise.
- Weeks 1-2Question map and baseline
Build the 20-question map. Test every question across ChatGPT, Perplexity and Google AI Overviews and log who gets cited today. Note which competitors keep appearing.
Output: A question map and a baseline citation log you can measure against. - Weeks 3-4Fix retrievability and pick the evidence
Make sure the relevant pages are crawlable, fast, and not hidden behind scripts. In parallel, pull the first-party data you are going to publish and check you can stand behind every figure.
Output: A clean technical base and a shortlist of original data points. - Weeks 5-10Build one asset a week
Work down the priority list. Each asset answers one question directly, carries one piece of first-party evidence, and links to the two or three nearest questions in the map.
Output: Six strong assets, published steadily rather than dumped. - Weeks 11-12Off-site push and re-test
Update directory listings, land two or three third-party mentions, and re-run the full question set. Compare against the baseline and reprioritise the next quarter.
Output: A second citation log, a movement read, and a refreshed priority list.
Where this goes wrong
- Chasing citation as the goal. Being named in an answer is a leading indicator, not revenue. If the citations climb and enquiries do not, you picked the wrong questions.
- Writing for models instead of buyers. Stuffed, robotic, over-structured pages read badly and convert worse. The same content has to work on a human who lands on it.
- Treating it as separate from the rest of marketing. Citation strategy works fastest alongside paid, email and outbound, because branded search and direct visits reinforce each other.
- Testing once and declaring failure. Answers move. One bad test in week two tells you nothing. Monthly, same questions, same method.
- Ignoring what the answer says about you. Sometimes you are cited but described wrong. Fixing the source page that caused it is usually a fast, high-value job.
"Citations are a scoreboard for visibility. Enquiries are the scoreboard for the business. Do not confuse the two."
Rather have someone build the pipeline while you run the business?
Book a 30 minute call. We will talk through where your leads come from now and what would move the number.
Read next
Common questions
If you want content and campaigns working together to bring in leads
- LinkedIn, Meta and Google campaigns run against a named target account list.
- Content and service pages built around the questions your buyers actually ask.
- Tracking and attribution wired up so you can see what is really driving enquiries.
- Creative built from your own material, shipped and tested every month.
- Reporting on cost per qualified pipeline rather than cost per click.
Starter £1,250/mo, Growth £1,725/mo, Scale £4,025+/mo. Content work sits inside the retainer where it is in scope. Anything outside that is quoted properly before it starts.