Most SEO teams are still optimizing the page. The LLMs only need the paragraph.
When ChatGPT, Perplexity, Claude, or Google’s AI Overviews answer a user’s question, they aren’t surfacing your URL — they’re lifting one or two sentences from somewhere on it and stitching that into the answer. Your H1, your meta description, your internal linking, your 2,400 words of context — none of it ships. A retrieval pass picks the chunk it thinks best matches the prompt, and the model paraphrases or quotes that chunk. Everything else on your page exists to support the chunk’s selection — not to be read.
That shift is the whole game. The work I do for clients now is half traditional on-page SEO and half something I’ve started calling citation engineering: deliberately constructing the small unit of prose the model is going to grab.
Why models pick the paragraphs they pick
Retrieval-augmented generation, which is what every consumer AI search product uses under the hood, breaks your page into chunks (typically 200–500 tokens), embeds each chunk, and ranks them against the user’s query. The top-ranked chunks get sent to the model along with the prompt. The model then either quotes them, paraphrases them, or — if the chunk is good enough to stand on its own — cites the source.
Three things make a chunk get picked:
1. Self-containment. It can be lifted out of the page and still make sense. No “as I mentioned above,” no “see the chart below,” no pronoun referring back to a prior paragraph.
2. Specificity. Concrete numbers, named entities, dates, mechanisms. Studies on AI citations have repeatedly shown that paragraphs containing a stat are roughly 22% more likely to be cited, and ones containing a direct quote are about 37% more likely. The models prefer evidence-shaped text.
3. Position. Roughly 44% of cited paragraphs come from the first 30% of a page. Bury the answer in the conclusion and it doesn’t get cited.
Combine those three and you have a citation unit — a paragraph engineered to be quoted in isolation, near the top, packed with claim-shaped sentences a model can reuse without having to think.
What this looks like in practice
I rewrite client pages around what I call “answer paragraphs.” Each one is 60–120 words, lives near the top of the page, and follows a tight pattern: open with the claim as a clean declarative sentence, follow with one or two sentences of supporting specificity (a stat, a named tool, a date, a mechanism), and close with a clarifying sentence that defines a term or qualifies the scope. No throat-clearing, no transitional phrases pointing elsewhere on the page.
The trick is that an answer paragraph also reads fine to humans. You’re not writing for two audiences. You’re writing one paragraph that satisfies a model’s retrieval logic and a busy buyer’s scan-pattern at the same time.
What to do this week
If you only have an hour:
- Pick your three highest-intent commercial pages. Read them aloud. The sentence a smart prospect would screenshot is your candidate citation unit. Move it into the first 30% of the page and tighten it to 60–120 self-contained words.
- Add one stat (with a real source) and one short quote (from a customer, an analyst, or a study) to each of those pages. Don’t bury them — put them inside the answer paragraph.
- Strip out the back-references. Anywhere you wrote “as we’ll see below” or “see Section 3,” rewrite the paragraph so it stands alone. LLMs hate dependency on context they don’t have.
- Run the page through ChatGPT or Perplexity yourself with a buyer-style query. If the model paraphrases your competitor instead of you, you have your before-state. Edit, redeploy, retest in a few days.
This is unglamorous work. It’s also the cheapest leverage available right now — the same content team you already have, asked to write tighter paragraphs in different positions.
Agencies: if your clients are starting to ask about AI SEO and you don’t have anyone in-house, Paris Roussos handles the work white-label — flat-rate, $500–$1,500/mo per end client, you keep the relationship. Email parisroussos@gmail.com for a sample audit.
The brand that gets cited becomes the default answer; the one that gets paraphrased loses the click and the credit. Engineer the paragraph or be the source someone else gets quoted over.