How to Rank in AI Search (GEO): 9 Tactics That Actually Get You Cited in 2026

Ranking in AI search is not the same job as ranking in Google. The tactics that get you cited by ChatGPT, Perplexity, and Google's AI Overviews are mostly about being retrievable and being quotable, not about backlinks or word count. This playbook lists the nine that have real evidence behind them, ranked by impact versus effort, and names the popular ones that measurably do nothing.

The reason this needs its own playbook is a single statistic. A peer-reviewed 2026 study from the University of Toronto compared what generative engines cite against Google's top ten results and found the overlap is tiny: about 4% for GPT-4o, 11% for Gemini, 13% for Claude, and 15% for Perplexity. AI engines surface a markedly different set of domains. So a page can rank on Google's first page and still be invisible to every AI assistant. AI visibility is its own workstream, and it is measured separately.

Key takeaways

• AI citation and Google ranking barely overlap (about 4% for GPT-4o), so treat AI visibility as a separate goal.
• The highest-leverage tactic is formatting: sections of 120 to 180 words that open with a direct answer earn about 70% more citations than short fragments.
• Word count does not help. Across 174,048 pages the correlation between length and citation was 0.04, effectively zero.
• ChatGPT Search runs on Bing's index, so being in Bing Webmaster Tools and pinging IndexNow is a real retrieval lever.
• llms.txt is a dead end. Named AI crawlers do not request it, Google ignores it, and no vendor confirms using it.
• Structured data is not required for AI features and does not change citation frequency. Keep it for Google rich results, not as a GEO play.

The short answer, by goal

What you want	The tactic with evidence	Effort
Get quoted inside an answer	Answer-first 120 to 180 word passages	Low
Be retrievable by ChatGPT	Bing Webmaster Tools plus IndexNow	Low
Be worth quoting	Publish original, cited statistics	Medium
Surface for brand-name prompts	Preferred Sources on Google	Low
Let AI read the page at all	Clean HTML plus AI crawlers allowed in robots.txt	Low
Stop wasting time	Skip llms.txt and AI-specific schema	Zero

1. Write answer-first passages (highest impact, lowest effort)

Lead every section with a direct "X is Y" answer in the first 40 to 60 words, and keep sections self-contained at 120 to 180 words. This is the single best-evidenced GEO tactic. SE Ranking's analysis of 216,524 pages found that sections of 120 to 180 words earned 4.6 citations on average, versus 2.7 for sections under 50 words, a 70% difference. AI systems extract passages, not whole articles, so a section that answers its own question in its opening sentence is the unit that gets lifted into an answer.

This is also the cheapest tactic, because it is formatting work on content you already have. Add a one-sentence answer capsule under each heading, a short takeaways block near the top, and a genuine question-and-answer section at the end. This very post is built that way on purpose.

2. Get into the index AI actually retrieves from

ChatGPT Search runs on Bing's index, so if you are not in Bing, you are not retrievable by ChatGPT, no matter how well you rank on Google. Roughly 87% of ChatGPT's cited results match Bing's top results. The fix is free and takes an afternoon: verify the site in Bing Webmaster Tools, submit the sitemap, and wire up IndexNow so new and updated URLs are pushed to Bing the moment they publish. Google's own systems handle Google AI Overviews and AI Mode, but Bing is the retrieval layer for the largest AI assistant, and almost no one optimizes for it.

3. Publish original data worth quoting

AI assistants preferentially cite specific, quotable statistics, so original numbers are citation bait. A Princeton study on generative engine optimization found that adding statistics, quotations, and cited sources measurably raised a page's visibility in LLM answers. First-party data is the strongest version of this because no competitor can copy it. Our own free-backlink survival study, where we hand-built 3,790 links and re-crawled every one months later, exists partly for this reason: a defensible number that other pages have to cite back to us.

4. Add "explore this post with AI" deep-links

Give readers one-tap buttons that open the post inside an AI assistant with a summarize prompt pre-filled. It is a light engagement feature, not a ranking signal, but it meets readers where research now happens and normalizes your content as an AI input. We ship it on every post on this blog:

Each button deep-links to Perplexity, ChatGPT, Grok, or Google's AI Mode with the current URL and a "summarize this post" prompt already filled in.

5. Become a Preferred Source on Google

Google's Preferred Sources lets a reader mark your site as one they want to see more of, and as of 27 May 2026 that preference now feeds AI Overviews and AI Mode, not just Top Stories. Adding a one-click button is optional but removes the friction of finding the setting. One rule matters: Preferred Sources works at the domain level only, so example.com qualifies but example.com/blog does not independently. Point the button at your root domain:

6. Let AI read the page: clean HTML and open crawlers

AI systems can only cite pages they can fetch and parse, so crawlability is a prerequisite, not an optimization. Two checks cover most of it. First, make sure the content is in the server-rendered HTML, not painted in by client-side JavaScript that a lightweight fetcher will miss. Second, decide crawler access deliberately in robots.txt, which is where AI access is actually negotiated today: GPTBot and OAI-SearchBot for OpenAI, PerplexityBot, and Google-Extended for Google's AI training. Blocking them removes you from those systems; allowing them is the price of being citable.

7. Make authorship real (E-E-A-T)

AI systems weigh the credibility signals around content, so a named author with a real bio and consistent identity beats an anonymous byline. Give every post a real author with a linked bio page, job title, and a consistent name used across the site and its structured data. This is the same experience-and-expertise signal Google rewards, and it costs nothing once set up.

8. Publish clean feeds

RSS and JSON feeds give machines a structured, low-noise way to ingest new content, which several AI and reader systems use. A valid feed is table-stakes plumbing rather than a growth hack, but it is cheap insurance and helps aggregators and assistants pick up new posts promptly.

9. Measure AI referrals separately

AI referral traffic will not show up cleanly in your normal reports, so build a dedicated view. In GA4, create a segment for referrals from chatgpt.com, perplexity.ai, gemini.google.com, and similar hosts. In Search Console, watch impressions on your AI-eligible pages. Because AI visibility overlaps so little with classic rankings, you cannot infer it from your Google position; you have to track it on its own.

The tactics that do not work

Three popular GEO tactics have real evidence against them. Spending time on these is the most common way to feel productive while moving nothing.

llms.txt does nothing. The idea is a machine-readable file telling AI models how to use your content. In practice, named AI crawlers do not request it. Ahrefs checked 137,210 domains and found 97% of llms.txt files received zero bot requests; SE Ranking analyzed 300,000 domains and found no statistically significant link to AI citations; and a 12-week server-log study saw GPTBot fetch robots.txt 3,990 times against 7 for llms.txt. We ran the same check on our own logs and found the identical pattern, documented in full here: AI crawlers hit robots.txt constantly and llms.txt essentially never. Google's Search Liaison has compared it to the long-abandoned keywords meta tag. Publish it once if you want the standards-track insurance; do not build anything around it.

AI-specific schema is not a citation lever. Google's own AI guidance states plainly that structured data is not required for generative AI features and there is no special markup to add. A study of 5.5 million AI responses found schema did not change citation frequency. Keep Article, Breadcrumb, Organization, and FAQ markup for their legitimate Google rich-result value, but do not sell schema internally as a GEO win.

Chasing word count backfires. Long content does not earn more citations. Across 174,048 pages the correlation between length and citation was 0.04, and 53% of cited pages were under 1,000 words. Completeness of the answer in a given passage beats total length every time.

The bottom line

Getting cited by AI is unglamorous and mostly free. Format answers so a machine can lift a clean passage, get into Bing so ChatGPT can retrieve you, publish numbers worth quoting, keep the page crawlable, and become a Preferred Source. Skip the tactics that only feel technical, llms.txt and AI-specific schema chief among them. And because AI cites a different web than Google ranks, measure it on its own. If your goal is the authority that underpins all of this, that still comes from real coverage and links, which is what our backlink gap audit is built to find.