SEO for Generative AI: An Operational Playbook for Getting Cited in AI Answers

Most teams treat generative AI visibility as a marketing opinion. It is not. It is a pipeline: a model retrieves a passage from your site, evaluates whether it is credible and self-contained, and decides whether to cite it. Every stage of that pipeline can break for operational reasons, and almost none of the breakages show up in a rank tracker. This playbook is a working process for SEO teams: fix access, audit which queries already trigger AI answers, restructure pages into extractable blocks, run quality control, and measure citations as their own channel.

SEO for Generative AI Is a Process Change, Not a New Channel

The temptation is to treat AI answers as a new marketing line item with a new budget and a new owner. In practice the work sits inside existing SEO operations: technical access checks, information architecture, editorial standards, and analytics segmentation. What changes is the unit of optimisation. Traditional SEO optimises a page for a query; generative retrieval optimises a passage for a retrieved context window. A page that ranks position four with one crisp definition paragraph can be cited, while the position-one page with the answer buried under 400 words of introduction may not be.

Assign one owner per stage below. In most organisations the technical owner is a developer or DevOps engineer, the content owner is an editor or SEO lead, and the measurement owner is an analyst. If a stage has no named owner, it will fail quietly.

Prerequisites: Fix Crawl Access and Machine-Readable Structure Before Writing Anything

Before any content work, confirm that AI crawlers and assistants can actually reach you. Three failure modes account for most invisibility: explicit robots.txt disallow rules copied from an old staging config, CDN or WAF bot rules that silently return 403 to unfamiliar user agents, and JavaScript-rendered content that never produces server-side HTML for a crawler to quote. Test each with curl and your WAF logs.

# Check whether a page is reachable and returns real HTML
curl -A "OAI-SearchBot/1.0" -I https://yourdomain.co.za/example-page
curl -A "Mozilla/5.0 (compatible; GPTBot/1.2)" https://yourdomain.co.za/example-page | head -c 500

# Inspect which bot rules your robots.txt exposes
curl https://yourdomain.co.za/robots.txt

Then add machine-readable structure: Organization and Article schema with author, datePublished and dateModified; Product schema on inventory pages; FAQPage only where questions and answers genuinely exist on the page. Submit new and updated URLs through IndexNow so Bing-fed assistants see changes quickly. Structured data does not guarantee citations, but it removes ambiguity about who wrote what and when.

Step 1: Audit Which Queries Already Trigger AI Answers You Could Be Cited In

Pull your top 200 non-branded queries by impressions from Search Console and manually check a sample of 40 to 60 in Google and in ChatGPT. Record which return an AI answer, whether a citation block appears, and who is cited. A simple spreadsheet with query, engine, answer present, cited domains, and whether you are cited is enough to start. You are looking for clusters where an AI answer exists, competitors are cited, and you have relevant content that is not appearing. Those clusters are your priority queue.

For a South African example: a Johannesburg-based B2B services firm targeting "outsourced payroll administration" queries competes in the same AI answers as UK and Australian providers. Geography does not isolate you; passage quality does.

Step 2: Restructure Pages Into Extractable, Answer-First Sections

Retrieval systems pull passages, so give them passages worth pulling. Each H2 section should open with a direct one to three sentence answer, then expand with evidence. Keep sections self-contained so a single block still makes sense if extracted. Convert comparisons and specifications into tables rather than prose paragraphs. Define terms once, plainly, near the top of the section. Avoid pronoun-heavy openings that only resolve when read in sequence.

Step 3: Add the Trust Signals Generative Engines Proxy For

Retrieval scoring favours content that looks like it came from a real, accountable source. Add author bylines with bio and role, show datePublished and dateModified visibly, cite primary sources outbound, and include first-hand data such as your own survey results or aggregate client benchmarks. Google's people-first guidance is the working standard here: write for a reader who wants an answer, not for a crawler counting keywords.

Step 4: Control Quality Before You Publish More Pages

Scaling content without quality control is the fastest way to lose citations. Before publishing, each page needs a factual check against source documents, a readability pass, a brand-voice pass, and a duplicate-intent check against the existing URL inventory. Straider, built from Prebo Digital's SEO work since 2017, codifies this with embedding-score gates, structured data enforcement, and human oversight before pages go live. See the platform for how that workflow runs.

Step 5: Measure AI Citations and Referrals Separately From Organic

Create a distinct analytics channel for referrals from chatgpt.com, perplexity.ai, copilot.microsoft.com, and gemini.google.com. Volumes are still smaller than classic organic in most accounts, but growth trends matter more than absolute numbers today. Separately, log citation appearances manually on your audit sample monthly. A ranking tool tells you position; it does not tell you whether your passage was quoted. Straider tracks visibility in Google AI Overviews and ChatGPT, giving a second read against your manual log.