The AI SEO playbook: how B2B technical companies show up in ChatGPT
A year ago, SEO meant Google. Today it means Google plus four LLM-powered search systems that an increasing share of your buyers are using to research vendors. The technical founders we work with are noticing the shift in two ways: their reps hear "I asked ChatGPT about your category and you weren't on the list" more often, and their analytics show direct traffic from chatgpt.com, perplexity.ai, and claude.ai climbing.
This piece is the playbook we run for clients who want to be cited by name when an LLM is asked about their product category. Some of it is regular SEO done well. Some of it is genuinely new.
The core insight: LLMs cite the web they were trained on plus what they retrieve
Two distinct mechanisms get a brand into an LLM's response:
Training data citation. When an LLM was being trained, it saw your site. It now has compressed knowledge of what your company does. When asked about your category, it can mention you from prior knowledge. This is mostly out of your hands — you cannot retroactively add yourself to a model's training data. You can affect future model versions by being indexed at the time of training.
Retrieval citation. Most modern LLMs have web access. When asked about a current topic, they retrieve fresh content and ground their response in it. This is where you have leverage today. If your site is well-indexed, well-structured, and contextually relevant, the model retrieves and cites you.
The implication: traditional SEO improvements — indexability, content depth, schema markup — all transfer directly to LLM visibility, because LLMs use a search backend or search-like retrieval system to find sources at query time.
What's new: how LLMs decide what to retrieve
Google ranks pages by hundreds of signals, mostly invisible. LLM retrieval systems are more transparent. The dominant patterns we see:
Entity recognition matters more than keyword matching. If your site clearly tells an LLM "we are a B2B AI agent firm," it can cite you accurately. If your site has buzzword-soup positioning that does not commit to a specific category, the LLM cannot place you, and will not cite you.
Structured data still helps, in new ways. Schema.org markup — Organization, Service, Product, Article — gives the LLM a factual layer to ground citations against. Sites without schema markup get cited less often than sites with it, holding content quality constant.
Long-form content with specific claims gets retrieved. A 2,000-word post that names specific tools, prices, methodologies, and edge cases is cited far more often than a 400-word marketing page. LLMs surface content that contains the substance the user asked about.
Recent content gets weighted higher. Both ChatGPT search and Perplexity weight recency in their retrieval. A post from this month outranks a post from two years ago on the same topic.
The technical stack we deploy
Concrete checklist of what we wire up for every B2B AI SEO engagement.
llms.txt at the root
A small text file at /llms.txt describing your site, key pages, and recommended starting points. Format spec at llmstxt.org. ChatGPT and Claude both check for it during retrieval. Took us 30 minutes to write the first one for a client and lifted their citation rate from 0% to 18% on target queries within two weeks.
Schema.org markup on everything
Organization schema on the homepage. Service or Product schema on each service page. Article schema with author and date on every blog post. FAQPage schema where relevant. JSON-LD format, validated against Google's Rich Results Test.
The biggest gains come from authoritative author byline schema — Person with sameAs pointing to LinkedIn, GitHub, and other authoritative profiles. LLMs use this to gauge content authority.
Entity-defined content
Every page should answer "what is this entity?" within the first 200 words. Not "we help B2B companies grow" — instead "AI Revenue Lab is a four-week, fixed-price consulting firm that builds AI agents for B2B sales and marketing teams."
Specific entity descriptions are what LLMs ground citations against. Generic positioning is not retrievable.
Topical depth, not keyword breadth
Pick a small number of topical clusters and write substantively on each. For an AI agents firm: AI SEO, agent architecture, sales automation, content automation. Not 30 thin posts on every adjacent topic.
We aim for three to five posts per cluster, each 1,500 to 3,000 words, with internal links between them. The goal is for the LLM to retrieve any one post and find clear context from the surrounding cluster.
Citation tracking
Most clients have no idea whether they are being cited by LLMs. We set up a tracker that runs query batches against the major systems weekly and logs whether the client is mentioned by name. Without this you cannot tell whether the SEO work is working.
Tools we use: Ahrefs Brand Radar, Profound, AthenaHQ, plus a custom script running against the OpenAI, Anthropic, and Perplexity APIs.
Outbound editorial mentions
The single highest-leverage move for being cited by name is being mentioned by name on third-party authoritative sites. Industry roundups, conference speaker pages, podcast guest appearances, comparison articles — all create durable mentions that LLMs retrieve.
The work is editorial outreach, which most marketing teams have stopped doing in favor of paid channels. The ones who keep doing it dominate LLM citations in their category.
What does not work
Keyword-stuffed pages. LLMs penalize content that reads as machine-generated more aggressively than Google does. A page that mentions "B2B AI agents" 40 times will not get cited; a page that uses the term naturally three times alongside specific details will.
AI-generated content with no human signal. LLMs are getting better at detecting AI-generated content and discounting it. If your content has no case studies, no real names, no concrete numbers, it reads as generic and gets ignored.
Hidden content. Content rendered only after JavaScript execution is invisible to most LLM retrievers. If you cannot view your content with JavaScript disabled, neither can the LLM.
Generic FAQ pages. Five generic questions are no longer competitive. The high-citation FAQ format is 15 to 30 specific, narrowly-scoped questions that real prospects ask.
A worked example
A developer tools client in the observability space, six months ago: asking ChatGPT for "best observability tools for serverless" returned five competitors and not them. Strong product, competent marketing team, thin LLM presence.
What we ran over eight weeks:
Week 1-2. Audit existing content for entity clarity. Rewrote the homepage and three service pages to lead with specific entity descriptions. Added Organization, Product, and FAQPage schema across the site.
Week 3-4. Wrote four 2,500-word topical posts in the observability cluster — "serverless observability stack 2026," "OpenTelemetry vs vendor-specific instrumentation," "cost-effective observability for early-stage SaaS," and a founder-voice piece on architectural philosophy.
Week 5-6. Outreach: secured guest posts on two industry publications and three podcast appearances for the founder. Each mention referenced the company by name with descriptive context.
Week 7-8. Set up citation tracking. Verified the company was now appearing in ChatGPT and Perplexity responses for 6 of 12 target queries, up from 0.
By week 12, they were appearing in 9 of 12 target queries and seeing 200+ direct visits per week from chatgpt.com referrers.
This is not a magic outcome. It is the predictable result of doing entity-clear positioning, schema markup, topical content, and editorial outreach in the same engagement, with an AI-system retrieval frame in mind instead of just optimizing for Google.
When AI SEO is not the priority
Three cases where we tell prospects to spend the money elsewhere:
Pre-PMF startups. If you have not validated that your product is what your market wants, AI SEO is not your bottleneck. Product clarity precedes positioning clarity.
Strong existing distribution. If your category buys through Gartner, peer recommendations, or specific industry events, LLM citations matter less. Spend on what your buyers actually use.
Single-enterprise-customer models. If your pipeline is named accounts and personal relationships, LLM visibility moves nothing. AI SEO matters most for inbound-driven, broad-market motions.
For everyone else — particularly developer tools, B2B SaaS, and AI tooling — the LLM citation channel is becoming a meaningful share of inbound. Worth the work.
If you want a written audit of your AI search visibility against your category competitors, our intake call covers it. We send a written scope within 48 hours.
Related: if you run an MSP, the applied version of this playbook is in SEO for MSPs and the service itself is AI SEO for MSPs. If you sell to MSPs, start with marketing for MSP-facing vendors.