The AI Search Playbook.
The whole method, no email gate.
The method I've used leading marketing at more than one B2B organisation here in New Zealand, written down in the order I actually do it. The point was never citations for their own sake. It was better leads. Use it yourself, hand it to your team, or hire me to run it. All three are fine by me.
To get a B2B brand found and cited by AI search: let AI crawlers in, tell machines who you are with schema and consistent brand facts, write answer-first pages with specifics a model can quote, build a connected web of content on the topics you want to own, earn mentions on sites AI already trusts, and measure citations, not just traffic.
Expect the fixes to ship in weeks and citations to build over three to six months. The ten chapters below cover each step, with checklists.
How AI search picks a source
Strip away the hype and every AI answer engine does roughly the same four things: it retrieves candidate pages from a search index, reads them, picks the passages that answer the question best, and decides how much to trust each source before it writes an answer.
Your job is to win at each step. Be in the index. Be easy to read. Have the clearest passage. Be a brand the model already has reason to believe.
| Step | What decides it | Chapter |
|---|---|---|
| Retrieve | Crawler access, indexing in Bing and Google, rendering | 03 |
| Read | Schema, headings, clean HTML, entity facts | 04 |
| Pick | Answer-first passages, specifics, structure | 05, 06 |
| Trust | Consistency, third-party mentions, track record | 07 |
Find out where you stand
Before you change anything, write down 20 to 50 questions your buyers genuinely ask. Not keywords. Questions. "What's the best X for a 50-person company?", "How much does X cost in NZ?", "X vs Y".
Run them through ChatGPT, Perplexity, Gemini and Google. Note who gets cited. That's your baseline, and it's usually humbling.
- List 20-50 real buyer questions (ask sales, not the SEO tool)
- Record who's cited for each, in each engine
- Run the free AI Visibility Check on your domain
- Turn on Bing Webmaster Tools and check its AI citation report
- Screenshot everything. You'll want the before picture in month four.
Let the bots in
You can't be cited if the AI can't read you. This is the most common problem I find, and the quickest to fix.
- robots.txt allows GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Bingbot and Googlebot
- Cloudflare or your firewall isn't silently blocking AI crawlers (check bot settings, not just robots.txt)
- Key content is in the HTML, not injected by JavaScript after load
- Your site is verified in both Google Search Console and Bing Webmaster Tools
- XML sitemap is clean and submitted; IndexNow is set up for Bing
- No accidental noindex tags or canonicals pointing at the wrong page
Tell machines who you are
A model needs to know you're a company that does a specific thing for a specific type of buyer, and that the same company shows up on LinkedIn, G2 and your own site. Schema markup and consistent facts are how you tell it.
- Organization schema on the homepage, with logo, description and sameAs links to your profiles
- Person schema for founders and authors; real bylines on articles
- Service or SoftwareApplication schema on what you sell
- Article and BreadcrumbList on content; FAQPage where there's genuine Q&A
- One brand description, used word-for-word on your site, LinkedIn, G2, Crunchbase and directories
- Validate everything in Google's Rich Results Test
Write answer-first pages
AI quotes passages, not pages. So every important page needs passages that make sense on their own and answer something directly.
The pattern is simple: answer in the first two sentences, then the detail. Use headings that are actual questions or clear statements, not "Our Approach". Put specifics in: prices, timelines, numbers, names. Vague copy doesn't get quoted, because there's nothing in it to quote.
- Each key page opens with a two-sentence answer to its main question
- Headings say what the section answers ("How much does X cost?" beats "Pricing philosophy")
- Specific numbers, prices and timelines, not "competitive" and "tailored"
- Tables for comparisons; lists for steps
- Titles and meta descriptions rewritten so each page clearly says what it's about
Build a content web
One great page doesn't make you an authority. A connected set of pages on a topic does. Pick the three or four topics you want to own, give each a hub page, and link every related article up to the hub and across to its neighbours.
- Three or four topics you want to own, chosen with sales
- One hub page per topic that explains it and links to everything underneath
- Every article links to its hub and two or three related pieces
- Menu and footer reflect those topics, not your org chart
- A glossary for your category's jargon
Get mentioned elsewhere
AI trusts brands that other trusted sites talk about. For B2B that usually means review sites (G2, Capterra), comparison and "best of" lists, industry publications, partner pages, podcasts, and yes, Reddit and LinkedIn.
You can't fake this, and you shouldn't try. You can make sure the mentions you already have are accurate, and work out which few missing ones matter most.
- Check which third-party sites AI is citing for your buyer questions
- Claim and update your profiles on the ones that matter
- Make sure every profile uses the same brand description
- Ask happy customers for reviews on the platforms AI is citing
- Show up where your buyers already talk. Answer questions, don't spam links.
Measure the right things
Traffic alone will lie to you. AI answers often satisfy the question without a click, so you need to measure citations and visibility too.
| Tool | What it tells you |
|---|---|
| Bing Webmaster Tools | How often your pages are cited in Copilot and Bing AI answers. Free. Underrated. |
| Google Search Console | Impressions and clicks, including AI Overviews and AI Mode, inside the normal performance report. |
| GA4 | Referral traffic from chatgpt.com, perplexity.ai and friends. Set up a channel group for it. |
| A prompt tracker | Your share of citations across your buyer questions in ChatGPT, Perplexity and Gemini, over time. |
| Your CRM | "How did you hear about us?" with an AI option. The number that matters most. |
Set a realistic timeline
The fixes ship fast. Citations take longer, because AI engines recrawl and rebuild trust on their own schedule.
- Weeks 1-4: foundations ship. Little visible change yet.
- Weeks 5-12: niche, long-tail questions start citing you, usually in Perplexity first.
- Months 3-6: mid- and head-term citations move. This is where it starts to compound.
- Months 6-12: content and mentions compound. Early movers tend to stay ahead.
Do it yourself, or hire someone?
Honestly? Plenty of teams can do most of this themselves. I've watched in-house teams in New Zealand do it well, once someone owned it. Everything above is the actual method. No secret chapter behind a form.
Do it yourself if you've got a developer with some spare time, someone in marketing who can own it for a few hours a week, and patience for the three-to-six-month curve.
Get help if nobody owns it, you've tried and it stalled, or you'd rather someone who's done it before made the calls and wrote the specs. That's what I do: AEO (answer engine optimisation) foundations from NZ$1,400 a month, or a $497 audit if you just want the to-do list and want to run it yourself.
Playbook FAQs.
Read it, run it,
or hand it to me.
If you'd rather someone else did all ten chapters, that's the job. If you want the to-do list for your own site, start with a $497 audit.