AI citation in 2026: what changed, and what to ship first.

On May 15, Google published official guidance saying "AI optimization" is just SEO — same content rules, same spam policy. ChatGPT, Perplexity, and Claude work differently. Here's how a multi-engine citation strategy looks now, and the four interventions that still move the needle.

The working definition

AI citation is the practice of making your brand more likely to be cited by the four major AI engines — ChatGPT, Claude, Gemini, and Perplexity — and to appear inside Google's AI Overviews when a user asks a question your brand can credibly answer. There are two different games inside that:

• For Google AI Overviews: it's SEO. The same content quality, structured data, and E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) signals that win regular Google rankings win the AI Overview too. Google has explicitly said: no separate "AEO" discipline, no AI-specific markup, no llms.txt needed.
• For ChatGPT, Perplexity, and Claude: the rules are different. These engines pull from a broader corpus, value clean structured data, read llms.txt, and lean heavily on third-party citations (G2, Capterra, Wikipedia, real press placements). This is where "AI citation" is its own discipline.

Together those two games are roughly 100% of AI search traffic in 2026. You have to play both.

Three things follow:

• You're optimizing for inclusion, not position. Either you're cited or you're not. There's no second place.
• The signals are mostly structural. Schema, Wikidata, clear paragraphs that lead with the answer, real third-party mentions. Style matters far less than people assume.
• The competitive set is broader. AI engines don't only pull from your top-10 SEO competitors. They pull from any source that has cleaner structured data and more credible third-party mentions than yours.

The five-axis scorecard

Our audit produces a single 0–100 number, but underneath it sits five axes. These are the ones that move month-over-month and that Google's policy update did not change.

1. Page-level structure

Whether the page text leads with the answer, uses skimmable paragraphs, marks H2/H3 cleanly, and avoids burying claims behind marketing language. AI engines do extractive reading; they reward structure. Google's Helpful Content System rewards the same thing.

2. Structured data

Organization, FAQPage, Article, Product, Service. JSON-LD with the right properties present and validated. Standard schema.org types only — Google's spam policy now discourages AI-specific markup extensions. Most sites have some schema — most schema is wrong.

3. E-E-A-T signals

Google's Experience, Expertise, Authoritativeness, Trustworthiness framework. Real author bylines with credentials, "about" and "contact" surfaces that prove a real business, original first-hand content rather than rewritten generics. This now drives both Google AI Overviews and ChatGPT's citation choices.

4. Identity reconciliation

Wikidata entity, llms.txt file (for ChatGPT and Perplexity), sameAs links to social, consistent name/address/phone across the open web. The reconciliation cascade decides which of two similar brands gets cited.

5. Citation density

How often your brand is mentioned in the corpus the engines actually retrieve from. G2 and Capterra reviews count. Real press placements count. Earned third-party rankings count — being placed IN a "Best X for Y" article on a real publication is worth more than publishing one yourself.

What to ship first (the four interventions)

1. Add an FAQPage block to your top-3 service pages

Eight to twelve question/answer pairs, marked up with FAQPage schema. Not "what is your company?" — actual questions buyers ask before they buy. Works for both Google AI Overviews and ChatGPT. This single intervention lifts citation rate measurably for ~80% of customers we audit.

2. Ship an llms.txt at your root (for ChatGPT and Perplexity)

One file, twelve lines, pointing AI crawlers at your canonical brand-fact pages. ChatGPT and Perplexity actively use it. Google says it's not necessary for them — so we don't pitch it as a Google play, we pitch it as ChatGPT/Perplexity coverage. Every fact in llms.txt must match the visible content on the linked pages (Google's spam policy targets sites that say one thing to bots and another to humans).

3. Clean your Wikidata entity

Add (or fix) instance-of, industry, official website, founded year, and Crunchbase ID. Five properties, fifteen minutes. The reconciliation cascade quietly does the rest. Every property must be sourced — Google's spam policy targets "inauthentic mentions," and a contested Wikidata edit hurts more than it helps.

4. Rewrite your top-5 paragraphs to lead with the answer

Most landing pages bury the answer two paragraphs in. Move it to the first sentence. AI engines do extractive reading; the answer needs to be where they look. This is also exactly what Google's Helpful Content System rewards — same intervention, two audiences.

What this looks like operationally

For a customer on our Growth plan, the first month maps roughly like this. Brand intake → audit → schema baseline → llms.txt → FAQ blocks on three landing pages → first batch of paragraph rewrites → first weekly digest summarizing what shipped. Approval queue running the whole time. Auto-approve off until you've watched the queue for two weeks.

For a regulated customer on Scale, slot the Compliance Queue in front of all of that. Every draft routes through it. SEC, FINRA, HIPAA, state-bar phrasing rules apply automatically.

You won't be able to feel the change for the first 14 days. The engines re-crawl on their own schedule. By day 30 our customers typically see citation count move from zero-or-low-single-digits to mid-single-digits across a 20-question vertical sample. By day 90 it tends to plateau at something more durable.

The most common mistake we see: customers want to ship a hundred new blog posts. The actual leverage is in twelve careful structural fixes, then a steady drip after.

What we're not claiming

AI citation is not a guarantee. The engines change. Your competitors will catch up. Citation count will move around month-to-month. The point of the discipline is to keep showing up faster than the field — not to "rank #1." We also don't use any tactic Google's May 2026 spam policy targeted: no hidden text, no fake mentions, no AI-only content. Everything we add to your site is visible, attributable, and built to survive a manual review.

If you want to see your starting point, the free audit takes about a minute and produces a public scorecard across all five axes. If you want us to fix it on a queue you control, the rest of the site explains how.