The short answer
There is no single ranking inside ChatGPT or Perplexity. Nobody is at position one. What actually happens is that the assistant runs a handful of searches, pulls back a shortlist of real web pages, throws most of them away, picks the sentences it can safely quote, and writes an answer from those. Your job is to be in that shortlist and to be the easiest thing on it to quote.
The good news for a small firm: when someone asks a narrow question, being famous matters less than being an obvious fit. "Best LinkedIn ads agency" is a fight you will lose. "LinkedIn ads agency for a 20 person engineering firm in the UK with a £2,000 a month budget" is a fight you can win, if your site says all of that out loud and can prove it.
"Do not try to be the answer to everything. Be the best evidenced answer to a small set of questions that are worth money to you."
What actually happens when someone asks
Told the normal way first.
- Someone types a full sentence into ChatGPT, with all their constraints in it.
- The assistant quietly turns that one sentence into six or seven different searches. This is the bit most people miss: you are not competing for their exact wording, you are competing across a whole family of searches around it ().
- It fetches a shortlist of real pages from a search index.
- It throws most of them out. Too vague, too old, does not really answer it.
- It pulls out the specific sentences it can lean on.
- It compares the two or three companies left against the constraints in the question.
- It writes the answer, and decides which sources to credit with a link.
Question ↓ understand the constraints ↓ fan out into related searches ↓ retrieve candidate pages ↓ rerank: quality, freshness, fit ↓ extract quotable passages ↓ compare the companies found ↓ generate the answer ↓ choose which sources to cite
Two things follow from this. First, getting pulled into the shortlist and getting credited in the answer are separate wins, and you want both. Second, none of this is a new channel with new rules. It runs on the same web index as normal search. Google says so plainly in its own guidance: the AI features sit on top of its existing ranking systems, and there is no special AI-only markup to add.
The eight questions your site has to pass
Everything in this guide comes back to these. Read them as yes/no questions about your own website.
The pages exist, load properly, and are in the search index.
Not just to their question, but to the six searches around it.
What you sell, who for, where, roughly what it costs.
Numbers, named clients, dates. Not adjectives.
The same claims exist off your own website.
When it compares three firms, you match their constraints best.
There is a clean sentence it can lift without risk.
Prices, people and results are not two years stale.
Most B2B sites fail on three and four. They are findable and vaguely relevant, and then they say "we deliver results for ambitious businesses" and lose to someone who wrote down an actual number.
Want more enquiries rather than a homework list?
Book a 30 minute call and we'll go through where your leads are actually leaking.
Score your own site
Twenty statements. Count the ones that are honestly true today. Under ten and you are invisible to this stuff. Ten to fifteen and you will show up occasionally. Sixteen or more and you will start getting named in answers for the narrow questions.
- Every page you want found loads without JavaScript problems and appears in Google.
- Your robots file does not block AI assistants by accident.
- You have a sitemap and it lists your real pages.
- Your important facts are in text on the page, not stuck inside an image or a PDF.
- One sentence describes your business the same way on every page.
- That sentence names what you sell, who you sell to, where, and roughly what it costs.
- You have a page saying who you are not for.
- Your prices, or at least a price range, are on the site in writing.
- At least five case studies with real numbers, on their own pages.
- Each one names the industry, the size of the company and the budget band.
- Your results pages carry dates.
- A named human is attached to the expertise, with a photo and a track record.
- Your company is described the same way on LinkedIn as on your site.
- At least three sites you do not own mention what you do and who for.
- You have reviews or testimonials that mention the specific work, not just 'great to work with'.
- Clients link to you from their own sites where that is natural.
- Your pages answer the question in the first paragraph, before the build-up.
- Headings are questions or plain statements, not clever titles.
- Numbers sit next to the claim they support.
- Nothing important is locked behind a form.
Nothing here is a trick. It is the same list a picky human buyer would work through, which is rather the point.
Work out what your buyers actually ask
Keywords were three words typed into a box. Questions are full sentences with constraints in them, and the constraints are where you win. Do not brainstorm these. Go and get the real ones.
- Your last 20 sales calls. The first two minutes, where they explain the problem in their own words. That wording is the gold.
- Your inbox. Enquiry forms and email threads. People type to you the way they type to an assistant.
- The assistants themselves. Ask ChatGPT and Perplexity your buyer's question and see who gets named and why.
- Where your buyers talk. Trade forums, LinkedIn comments, industry groups. Not marketing forums.
Then sort them. A question is worth writing about when someone with a budget asks it close to buying. "What is B2B marketing" is not that. "How much should a 20 person engineering firm spend on LinkedIn ads" is.
One row per question. A spreadsheet is fine, this is not a system to buy.
Question (their words, not yours)
Who asks it → job title, company size
How close to buying → early / comparing / ready
Worth money? → yes / no
Do we answer it? → page URL, or blank
Proof we can use → case study, number, screenshot
Answered in → one sentence, top of page
Fill twenty rows before you write anything. The gaps in the 'do we answer it' column are your content plan for the next quarter.
Proof is the whole game
When an assistant compares three suppliers, it is looking for evidence that one of them matches the constraints in the question. Adjectives are not evidence. This is:
"We deliver exceptional results for B2B clients across a range of sectors."
"A machinery manufacturer, 30 staff, UK. Cost per lead fell from over £300 to £55. 17 new clients worth £26k each, from under £250k of ad spend over five years."
Every case study wants its own page with its own address. Industry, company size, budget band, timeframe, and the numbers, all in text. A carousel on your homepage is invisible here. So is a PDF.
Get other sites saying it too
A claim only on your own website is a claim. The same claim on three other sites is a fact, as far as these systems are concerned (). This is the most neglected lever in B2B and the hardest to fake, which is exactly why it counts.
- Client sites and partner pages describing the work you did for them.
- Reviews that mention the specific thing: the sector, the budget, the outcome.
- Trade press and industry association listings in your clients' world, not yours.
- Podcast appearances and guest pieces where a named person says something specific.
- Your LinkedIn company page, worded the same as your website.
Topical beats big. One mention in a trade magazine your buyers read is worth more than ten generic directory listings.
The first 90 days
In order, because the order matters. There is no point publishing brilliant answers on a site that cannot explain what your company is.
- Days 1–15Make sure you are readable
Check your pages appear in Google. Check your robots file is not blocking AI assistants. Get your key facts out of images and PDFs and into text. Write the one sentence that describes your business and put it everywhere, identically.
Output: A site that can be found and understood, and one agreed sentence. - Days 16–35Publish the proof
Five case studies on their own pages, each with industry, company size, budget band, timeframe and real numbers. Add prices or a price range. Add a page saying who you are not for.
Output: Five evidence pages and an honest fit page. - Days 36–65Answer the money questions
Take the top five rows from your question map and write the page for each. Answer in the first paragraph, then earn the rest. One number or example per section.
Output: Five pages aimed at people close to buying. - Days 45–80Get it said elsewhere
Ask three happy clients for a specific review. Ask two for a link or a mention. Line up one guest piece or podcast in your buyers' industry, not in marketing.
Output: Five or six independent mentions saying the same thing your site says. - Days 66–90Measure and repeat
Run your fixed list of buyer questions through the assistants, log who gets named, compare against your first run. Fix the questions where you are still absent.
Output: A monthly habit and a number that can go up.
How to tell if it is working
There is no ranking report for this, so you build your own. It takes about an hour a month.
- Fix a list of 30 questions from your question map. Same list every month, or the numbers mean nothing.
- Run them through ChatGPT, Perplexity and Google's AI answers. Log whether you were mentioned and whether you were linked.
- Watch for traffic from the assistants in your analytics. It is small, and it converts unusually well, because the assistant has already pre-sold you.
- Watch searches for your company name. When people hear about you in an answer and then go and look you up, this is where it shows.
- Ask new enquiries how they found you. Blunt, unfashionable, and still the most reliable signal you have.
Bing's Webmaster Tools now shows which AI systems cited your pages and the searches they used to find them. It is free and it is the only place you get that view, so connect it even if nobody you know uses Bing.
"Citations are not the goal. Enquiries are. A hundred mentions on questions nobody buys from is a hobby, not a channel."
What not to do
- Mass-produce 500 "AI-optimised" pages. Thin pages that repeat each other get filtered out and drag the good pages down with them.
- Add an llms.txt file and expect results. It is a proposed convention, not a ranking mechanism, and the major assistants do not currently use it to decide anything. Harmless, just not the job.
- Hunt for special AI markup. There isn't any. Google has said so directly. Standard for your organisation, your articles and your services is the whole list.
- Write content that sounds like AI wrote it. Generic filler gives an assistant nothing specific to quote, which is the one thing you needed.
- Obsess over one chatbot. Your buyers use whatever is on their phone. Build evidence once and it works across all of them.
- Treat it as separate from SEO. It runs on the same index. If your site is a mess in normal search, it is a mess here.
Go deeper on any part of this
The starting point, in more detail.
The full scorecard, run as an exercise.
Page-by-page layout.
The five pillars over 90 days.
The four types worth adding.
Where to mine them and how to score them.
Common questions
Where this comes from
The platforms publish more than people realise. Everything above is either from their own documentation or from running campaigns and content for B2B clients ourselves. Where something is a judgement call rather than a documented fact, we have said so on the page.
- Google — Optimising your website for generative AI features
- Google — AI features and your website (query fan-out)
- OpenAI — Searching the web with ChatGPT
- OpenAI — Publishers and developers FAQ
- Bing — AI Performance in Webmaster Tools
Written August 2026. The platforms change this stuff regularly, so treat anything that looks like a specific mechanism as true at the time of writing rather than permanent.
If what you actually want is more qualified enquiries
- LinkedIn, Meta and Google campaigns run against a named list of target accounts.
- Service pages and content built around the questions your buyers really ask.
- Tracking and attribution set up so you can see what is genuinely driving enquiries.
- Ad creative built from your own material, shipped and tested every month.
- Reporting on cost per qualified pipeline, not cost per click.
Starter £1,250/mo, Growth £1,725/mo, Scale £4,025+/mo. Content and positioning work sits inside the retainer where it is in scope. Anything beyond that gets quoted properly before it starts.