Back

AI Search Traffic Is Up 527%. Here's How to Get Cited.

·8 min read·en

AI · Marketing

Your SEO is working. You're ranking on page one. But something has changed: users are skipping your link entirely and asking ChatGPT instead. ChatGPT answers their question. The citation goes to someone else.

This isn't hypothetical. AI search referral traffic grew 527% in the first five months of 2025. At the same time, click-through rates on traditional search results dropped as much as 58% due to AI Overviews. If your content isn't getting cited by AI engines, you're missing the fastest-growing traffic channel right now.

This is the playbook for GEO — Generative Engine Optimization.

How AI Decides What to Cite

Every AI search engine follows roughly the same pipeline: user asks a question → AI breaks it into sub-queries → retrieves candidate pages → extracts the best passages → assembles an answer with citations.

There are four points where you can intervene: getting indexed (visibility), getting your passages selected (structure), getting accurately paraphrased (precision), and being considered trustworthy enough to cite (authority). Everything below is about one of these four things.

Step 1: Technical Foundation (30 Minutes)

The highest ROI work, and the most commonly skipped.

Allow AI crawlers in robots.txt

Make sure none of these are blocked:

  • OAI-SearchBot (ChatGPT search — most important)
  • ChatGPT-User (ChatGPT browse mode)
  • PerplexityBot
  • Claude-User (user-triggered fetches)
  • Google-Extended (Gemini / AI Overview)
  • Bingbot (Copilot)

One useful nuance: you can block ClaudeBot (used for training data collection) while allowing Claude-User (used when actual users ask Claude to look something up). You're opting out of training while staying in the citation pool.

Deploy llms.txt

Introduced in late 2024, this is essentially a structured menu of your site for AI systems. Drop it in your root directory as plain Markdown:

# Your Brand
> One sentence: who you are and what you do.

## Core Products
- [Product A](URL): brief description

## Documentation
- [Technical Docs](URL): description

If you have extensive documentation, consider an llms-full.txt that consolidates everything into one Markdown file. Models like Claude reason better over unified datasets than scattered HTML pages.

Add Schema markup (JSON-LD)

AI engines index entities — people, organizations, concepts — not just keywords. Schema is how you formally declare what you are. Pages with proper Schema markup get 30-40% better visibility in AI results.

Prioritize: Article, FAQPage (Perplexity loves this), HowTo, and Person with sameAs linking the author to LinkedIn or GitHub.

Ensure server-side rendering

Critical content needs to be in the initial HTML response. AI crawlers are not good at executing JavaScript-heavy client-side rendering. Target under 2 seconds load time — Copilot has a documented threshold.

Step 2: Content Structure

This is where the leverage is. Princeton's GEO study (KDD 2024) ran controlled tests on nine optimization techniques. The results were clear: citing authoritative sources (+40% citation rate), adding statistics (+37%), and including expert quotes (+30%) were the biggest movers. Keyword stuffing actually decreased citation rates by 10%.

Four rules for AI-citable writing:

Put the conclusion first. 44.2% of AI citations come from the first third of an article. Open every important page with a 50-100 word TL;DR that gives the answer directly. The first sentence of every section should be the takeaway. The optimal passage length for AI extraction is 40-60 words — one idea per paragraph.

Write headers as questions. 78.4% of question-related citations reference the heading itself. Rewrite your H2s and H3s as questions your readers would actually type: "What is X?" "How does X work?" "X vs Y — what's the difference?" This directly mirrors how AI systems decompose user queries.

Make each paragraph self-contained. Every paragraph should be citable without the surrounding context. Avoid phrases like "as mentioned above" — they break when extracted. AI pulls passage-level chunks, not whole articles.

Use definitive language. Text using "X is" or "X refers to" gets cited at roughly twice the rate of hedged or vague phrasing. Aim for the tone of a good analyst note: factual, specific, a mix of data and interpretation. Not a sales pitch, not a research paper.

Content types that get cited, ranked:

Comparison articles (X vs Y) account for roughly 33% of citations — by far the most citable format. Authority guides (15%), original research and data (12%), and ranked lists (10%) follow. What AI won't cite: generic posts without structure, marketing fluff, paywalled content, anything without a date or author.

Step 3: Platform-Specific Strategy

Each AI engine has different backends and different preferences. Optimizing without knowing which is which means wasted effort.

ChatGPT runs on Bing's index. The most important insight: content that looks like how ChatGPT answers questions gets cited at much higher rates — this structural similarity accounts for 55% of citation weight, dwarfing domain authority (12%). Study how ChatGPT formats responses, then write your content that way. Freshness matters a lot: content updated within 30 days gets cited 3.2x more often. Anything you're competing for should be refreshed at least monthly.

Perplexity draws from its own index plus Google. The biggest differentiator is FAQPage Schema and public PDFs. If you have white papers or research reports, make them publicly downloadable — Perplexity strongly prefers citing PDFs. It also rewards high publishing frequency and tight, self-contained paragraphs.

Claude uses Brave Search as its backend — not Google, not Bing. That means you need to verify your content is indexed at search.brave.com before anything else. Claude is highly selective: it processes a lot but cites sparingly, favoring the most factually accurate sources. Optimize for information density: specific numbers, named sources, dated statistics.

Gemini / AI Overview runs on Google's index and Knowledge Graph. Its distinctive behavior is "fan-out queries" — it automatically decomposes a question into multiple sub-queries. You want topical depth: a cluster of content covering a subject from multiple angles, with FAQ sections addressing each sub-question. Schema markup has the highest leverage here (+30-40%).

Copilot is pure Bing, with extra weight given to LinkedIn and GitHub content. Submit through Bing Webmaster Tools and use the IndexNow protocol for faster indexing.

Grok has exclusive access to the full X (Twitter) firehose. Reply quality and depth account for over 75% of post ranking — a substantive reply outweighs a like by 50x. One trap: putting external links in the main post drops your score by about 50%. Links should go in replies only.

Step 4: Build Third-Party Presence

Here's the number most people miss: brands get cited through third-party sources at 6.5x the rate of their own websites.

Optimizing only your own site is playing half the game. You need a presence where AI engines already look:

  • Wikipedia: 7.8% of ChatGPT citations. Make sure your entry is accurate and current.
  • Reddit: 33% of AI citations come from here. OpenAI has a direct data partnership with Reddit. Community consensus is one of ChatGPT's primary citation signals.
  • Review sites (G2, Capterra): Perplexity relies heavily on these for product comparisons.
  • YouTube: High citation rate in Google AI Overviews — upload proper subtitle files.

On Reddit, the play is genuine participation in relevant communities, not promotion. You want your brand to appear organically in high-karma threads, in discussions where it actually belongs. That's what gets pulled into AI responses.

Step 5: Measure and Iterate

The minimum viable monitoring loop is weekly. Maintain 30-80 fixed prompts covering brand definitions, competitor comparisons, and use case questions. Run them through ChatGPT, Perplexity, and Google. Track: are you mentioned? Is there a citation with a link? Which URL? Is the description accurate?

Useful prompt templates:

  • "What is [your brand] and who is it best for? Include sources."
  • "[Your brand] vs [competitor] — when should I choose each? Cite sources."

In Google Analytics, watch utm_source=chatgpt.com (OpenAI appends this automatically) and referral traffic from Perplexity and other AI domains.

Where to Start

If you have a weekend, here's the priority order:

Day one — technical foundation: Update robots.txt, deploy llms.txt, add Schema to your core pages. One-time work, two hours maximum, but without it everything else gets a 30-40% haircut.

Day two — content overhaul: Pick your three to five most important pages. Add a TL;DR block at the top. Rewrite headers as questions. Audit each paragraph for self-containment. Add statistics and named sources anywhere you can.

Ongoing: Monthly content refreshes on competitive topics. Consistent third-party presence building. Weekly prompt monitoring.

GEO is a long game, same as SEO. The difference is that most people haven't started yet — and compounding works best when you're early.

Home
About
Projects
Blog