The five minute version
For a niche B2B firm with a handful of decent pages and no idea whether an assistant has ever mentioned them.
- 1Write down the twenty questions your buyers actually ask you on a first call. Those are the prompts.
- 2Run all twenty through ChatGPT, Perplexity and Google and log who gets named. That is your real competitive set.
- 3Pick the five questions you can answer better than anyone, and answer each one in the first hundred words of a page.
- 4Put one thing in each page nobody can restate: your own numbers, your own named method, a worked example with real figures.
- 5Get named off-site as well. Roundups, community threads, podcasts with transcripts. Your own site is only one input.
- 6Re-run the twenty prompts every month and watch the trend, not the single result.
The short answer
Answer engines cite pages that clearly answer a specific question, come from a source that looks credible on that subject, and can be quoted in a couple of sentences without the model having to interpret you. For a niche B2B firm that means fewer, sharper pages on the exact questions your buyers ask, containing something a competitor cannot copy: your own numbers, your own method, your own worked examples.
It is not a separate discipline from SEO. Good search fundamentals still carry most of the weight. What changes is the unit you optimise for. You are no longer trying to occupy a slot on a page of ten blue links. You are trying to be the sentence a model reaches for.
How an answer engine picks who to cite
The mechanics differ across ChatGPT search, Perplexity and Google's AI Overviews, and none of them publish the full recipe. But the shape is consistent enough to work with. A question comes in. It gets turned into one or more searches. A set of candidate pages comes back from an underlying index, usually Google or Bing. The model reads chunks of those pages, drafts an answer, and attaches links to the chunks it leaned on.
Two things follow from that. First, if you are not retrievable in conventional search you are not a candidate at all, so the fundamentals are not optional. Second, retrieval happens at the chunk level, not the page level. A page that ranks well but buries the answer under 600 words of preamble is a worse citation candidate than a leaner page that states the answer in its second paragraph.
"You are not competing for a position any more. You are competing to be the paragraph that gets lifted."
There is a third factor that gets underrated: models weigh what other sources say about you. Being repeatedly named alongside your category in places the model already trusts does more than another page on your own site. That is closer to PR than to SEO, and most B2B firms are doing none of it.
Find the questions worth being the answer to
Keyword volume is a poor guide here. Prompts are long, conversational and often have almost no search volume attached to them, which is exactly why they are winnable. A question asked by fifty of the right people a month is worth more to a niche firm than a head term with ten thousand searches and no buying intent.
- Mine your own inbox. The questions prospects actually ask on calls, in emails and in the first ten minutes of a discovery conversation. Those are the prompts, almost word for word.
- Ask the engines directly. Put twenty buyer questions into ChatGPT, Perplexity and Google and record who gets cited. That list is your real competitive set, and it is often not who you assumed.
- Prioritise decision questions over definition questions. "What does LinkedIn Ads management cost in the UK" beats "what is LinkedIn advertising". Definitions are already covered by the platforms themselves.
- Take the questions with a defensible answer. If your answer would be identical to everyone else's, being cited gains you nothing.
Swap in your service and your sector, run the set monthly, log every firm that gets named.
1. Who are the best {service} agencies for {sector} in the UK?
2. How much should a {sector} firm budget for {service}?
3. Is {service} worth it for a company with a small team?
4. Why are my {service} campaigns getting clicks but no enquiries?
5. What are realistic benchmarks for {service} in {sector}?
6. How do I prove {service} influenced a deal in my CRM?
7. What should I ask an agency before signing a {service} retainer?
8. In-house or agency for {service}, and at what budget does that flip?
Record three columns each month: the sources cited, whether you were named, and which page of yours was quoted if you were. Ten minutes a month, and it is the only honest scoreboard you have.
Write pages a model can lift from
Structure is doing more work than word count. The goal is that any 200-word window of your page makes sense on its own, because that is roughly how it will be read.
- Answer in the first hundred words. A short direct answer near the top, before the context, the caveats and the story. Everything else can follow it.
- Use headings that are questions. They map to prompts and they make the retrieval job obvious.
- Keep sections self-contained. Avoid "as we covered above". A chunk pulled out of sequence should still stand up.
- Put comparisons in tables and numbers in plain text. Figures baked into images are invisible. State the number in the sentence as well.
- Add a real FAQ block. Question-and-answer pairs are the easiest thing in the world to quote, which is precisely why they get quoted.
- Date the page and say when it was last reviewed. Recency is a tiebreaker on anything that changes year to year.
"In today's competitive B2B landscape, advertising costs can vary significantly depending on a wide range of factors, and it is important to consider your unique business goals before deciding on a budget."
No number, no position, nothing to attribute. A model can write that sentence itself.
"Most UK B2B firms running LinkedIn Ads properly are working with a monthly media budget starting around £1,000 to £2,000, plus management. Below that, you are testing, not scaling."
A figure, a threshold and a stated view. That needs a source, and the source is you.
Read your own page back and ask one question of every section: could a model quote this paragraph on its own and be saying something? If the honest answer is no, the paragraph is filler and it is diluting the chunks either side of it.
FAQPage and Article structured data is worth adding because it removes ambiguity for the crawlers feeding the index. Treat it as hygiene rather than a lever. Nobody gets cited because of schema alone.
Want more leads coming in, not just better-looking content?
Book a 30 minute call. Tell us where your pipeline is thin and we will give you a straight answer on whether we can help.
Give it something only you have
This is the part that separates firms who get cited from firms who publish. Models synthesise. When ten pages say the same thing in different words, there is no reason to name any of them. When one page carries a number nobody else has, it gets attributed, because the claim needs a source.
- Publish your own benchmarks. Anonymised, aggregated, stated with the sample size and the period. Our UK LinkedIn benchmark work gets referenced precisely because it is ours and it is specific.
- Show worked examples with real figures. Spend, CPC, what changed, what happened next. Vague case studies are unquotable.
- Name your method. A framework with a name and a definition is easy for a model to attribute. An unnamed process is not.
- Take a clear position. Hedged writing is the least citable writing there is. Say what you think and say why.
The obvious guardrail: none of this works if the numbers are invented. Fabricated stats get contradicted by other sources, and a model that finds a conflict tends to cite neither. Only publish figures you could defend line by line if a client asked.
Get named in the places models read
Your own site is one input. The rest of the answer is assembled from sources the model already trusts, and for B2B that skews towards forums, industry publications, directories, comparison sites and YouTube transcripts more than most people expect.
- Be a real participant where your buyers talk. Reddit and niche communities get retrieved heavily. Turn up as a practitioner answering questions, not as a link drop, or you will get removed and it will have cost you time for nothing.
- Get into the roundups. "Best LinkedIn Ads agencies UK" style lists are exactly the pages models read when someone asks for a recommendation. Being absent from all of them is a choice.
- Do guest pieces and podcasts with transcripts. A transcribed conversation is a large body of attributed text with your name and your category in it.
- Keep your entity details consistent. Same company name, same description, same founder name everywhere. Inconsistency makes it harder for anything to associate you with your specialism.
- Publish on YouTube with proper descriptions. Transcripts are indexed and quoted, and video is the cheapest way to produce a lot of defensible first-person material.
Fix the technical bits that block retrieval
Most B2B sites do not have an exotic problem here. They have one or two boring ones that quietly exclude them.
- Serve content without JavaScript execution. If the page is empty until a script runs, some crawlers get nothing. Prerendered or server-rendered HTML is the safe position.
- Decide on the AI crawlers deliberately. GPTBot, PerplexityBot, ClaudeBot and the rest are controlled in robots.txt. Blocking them is a legitimate choice, but do not block them by accident and then wonder why you are never cited.
- Unique titles and descriptions on every page. Duplicated metadata makes pages look interchangeable to anything ranking them.
- Keep the answer out of gated content. A PDF behind a form cannot be quoted. Publish the substance and gate the tool or the template instead.
- Make sure the page loads quickly and cleanly. Same reasons as always. Nothing has changed there.
The first 30 days, in order
Nobody has a spare quarter to rebuild a site. This is the sequence I would run if you gave me a month and one afternoon a week.
- Week 1Baseline before you touch anything
Build the twenty question prompt panel from your call notes, run it across ChatGPT, Perplexity and Google, and log who is named. Check robots.txt while you are there in case you are blocking the crawlers by accident.
Output: A scoreboard and a list of the firms currently getting your answer. - Week 2Fix the two pages you already have
Take your best existing pages on buying questions and move the answer to the top. Question headings, self contained sections, numbers written in text rather than sat inside an image, a real FAQ block at the bottom.
Output: Two pages that can be lifted from without interpretation. - Week 3Publish the thing only you have
One page carrying your own figures: anonymised benchmarks with the sample size and period, or a worked example with spend, cost per click, what you changed and what happened. This is the page that earns attribution rather than paraphrase.
Output: A defensible source other people have to cite. - Week 4Get named somewhere that is not your site
Answer three real questions in the communities your buyers use as a practitioner, pitch one guest piece or podcast, and get yourself considered for the two roundups that already rank for your category.
Output: Off-site mentions in sources the models already read.
Then stop and re-run the panel. One month will rarely move a citation, but it will tell you whether the pages are even being retrieved, which is the thing worth knowing before you spend a quarter on it.
How to tell if any of it is working
There is no Search Console for answer engines, and anyone selling you a precise citation share number is estimating. Three measurements are worth keeping, and none of them is perfect.
| What to track | How | What it tells you |
|---|---|---|
| Prompt panel | Run the same 20 to 30 buyer questions monthly, log who is cited. | Movement over time. Answers vary run to run, so read the trend not one result. |
| Referral traffic | GA4 sources containing chatgpt, perplexity, copilot, gemini. | Real clicks arriving from assistants. Small numbers, usually high intent. |
| Self-reported source | One open field on the enquiry form. | The only place "ChatGPT suggested you" ever shows up. Crude and genuinely useful. |
Expect the traffic numbers to look tiny next to organic search. Judge them on what they turn into rather than volume. An assistant that recommends you by name has done a chunk of the qualifying job before the person ever lands on your site.
What is not worth your time
- llms.txt. A proposed standard the major assistants do not commit to using. It costs ten minutes, so add it if you like, but it is not a strategy.
- Rewriting your whole site for "AI readability". Fix the pages that answer buying questions. Leave the rest alone.
- Volume publishing. Forty thin pages give a model forty reasons to cite someone with more authority. Ten strong ones give it a reason to cite you.
- Citation-tracking tools bought before you have anything to track. Do the manual prompt panel for a quarter first. If it becomes a real channel, then pay for monitoring.
- Chasing head terms. Ask an assistant a broad category question and it names the biggest brands in the market. That fight is not winnable and it is not where your buyer is anyway.
"Nobody wins this by publishing more. You win it by being the only firm in your niche that has bothered to write down what it actually knows."
Rather have someone turn this into pipeline for you?
Book a 30 minute call and tell us what your sales team needs more of. We will tell you honestly whether we are the right fit.
Read next
Common questions
Where this comes from
Most writing on this subject is guesswork repeated confidently. These are the primary sources from the platforms themselves. Where something below is inference rather than documented, we have said so in the text.
- Google — Optimising your website for generative AI features
- Google — AI features and your website
- OpenAI — Searching the web with ChatGPT
- OpenAI — Publishers and developers FAQ
- Bing — AI Performance in Webmaster Tools
Written August 2026. The platforms change this regularly, so treat any specific mechanism as true at the time of writing rather than permanent.
If you want leads coming in, not just mentions
- LinkedIn, Meta and Google campaigns run against a named target account list.
- Creative built from your own material, shipped and tested every month.
- Tracking and attribution wired up so you can see what is actually working.
- Content and landing pages written to answer the questions your buyers ask.
- 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.