How do I get cited by Perplexity or Google AI Overviews?

You cannot guarantee a citation. You can write self-contained answers, original evidence, and clear headings so retrieval systems have something to lift. See our GEO overview for the strategy; this post is the paragraph templates.

What is a quotable span?

A sentence or short block that still makes sense with no surrounding context: a definition, a numbered step, or a dated statistic with a method. Throat-clearing intros rarely get quoted.

Should I humanize before or after writing the extractable blocks?

After. Lock the answer, the number, and the source. Then one Human Writes pass on stiff prose. Humanizing empty SEO filler produces polished filler that still will not be cited.

Is this the same as the GEO guide?

No. The GEO guide covers what answer engines reward and the overall workflow. This post is how to format the three block types those engines actually lift.

Write Pages AI Overviews and Perplexity Can Cite

Answer engines do not cite "your page." They lift spans: a definition, a step list, a number with a method. Most AI-drafted blogs never write those spans. They write history, hedging, and "it depends," then hope the model will invent a pull quote.

This is the paragraph-level follow-on to Generative Engine Optimization (GEO). Not another GEO-versus-SEO lecture. Not another E-E-A-T sermon (that lives here). This is how to write a block an overview can copy.

You still cannot guarantee ChatGPT, Perplexity, or AI Overviews will pick you. You can stop writing unquotable prose.

What gets lifted (and what does not)

Engine-friendly	Usually ignored
One-sentence definition that names the term	"In today's rapidly evolving landscape…"
Numbered steps that stand alone	A 400-word anecdote before the how-to
Dated first-party stat with method	"Studies show" with no study
Comparison table with a winner and a limit	Keyword-stuffed H2s with no answer
FAQ answers that repeat the question in the first line	"As mentioned above" fragments

Write at least three extractable blocks per guide: definition, steps, proof. Then humanize rhythm around them. Do not humanize the number into "approximately significant improvement."

Template 1: definition box

Put this under the H1 or as the first H2. One or two sentences. Term in the first five words. No throat-clearing.

Pattern:

[Term] is [plain definition]. It [does X] by [mechanism], not by [common confusion].

Weak (unliftable):

When we talk about citation in the age of AI, it is important to consider many factors.

Strong:

A quotable span is a self-contained sentence or short block that still makes sense if an answer engine copies it with no surrounding paragraph.

After the box, add one sentence of nuance. Engines often take the box and skip the nuance. That is fine. Humans need the nuance so you do not overclaim.

Template 2: numbered steps

Use a real numbered list, not a paragraph that says "first, then, finally." Each step should be a complete instruction.

Pattern for each step:

Verb + object. One constraint. One "don't."

Weak:

There are several things to keep in mind when optimizing for generative engines.

Strong:

State the user question in the H1. Phrase it the way someone types into ChatGPT, not as a clever brand line.
Answer in the first 150 words. Definition or recommendation before history.
Place one original number on the page. Date it. Name the method. Link the source or say it is first-party.
Write FAQ answers as standalone sentences. Repeat the question's key noun in the first line so the chunk retrieves.

If a step needs your product, say so after the instruction. Do not hide a pitch inside step 1.

Template 3: original stat

Engines prefer a number they can attribute. Invented percentages are worse than no percentage. Same discipline as citation best practices.

Pattern:

In [month year], we [method, n=]. [Result]. That does not prove [overclaim].

Weak:

Most marketers struggle with AI Overviews.

Strong:

In a July 2026 internal review of 12 Human Writes blog URLs, pages that led with a one-sentence definition were the ones we could find quoted in reproducible Perplexity answers. Twelve URLs is not a ranking study. It is a reason we still write the box.

If you do not have a first-party number, quote a named primary source with a date. Do not let ChatGPT supply "research shows 73%."

Where Human Writes helps

Block	Humanize?	Why
Definition sentence	Light pass	Keep the term and the mechanism
Numbered steps	After the verbs are right	Cut "leverage" and even clause length
Stat + method	No, except grammar	Do not round or "smooth" the figure
Setup paragraphs around the blocks	Yes, one pass	This is where AI sounds like every other GEO post

Paste stiff setup into Human Writes once. Do not run the stat through a rewriter until it is no longer checkable.

Page skeleton (copy this)

H1 = the question
Definition box (template 1)
Who this is for / who it is not for (two bullets)
Numbered method (template 2)
Proof (template 3) or a comparison table
Limits and "you cannot guarantee citation"
FAQ that repeats the question in the answer

Then one voice pass. Then publish. Update the date when the proof changes.

What not to do

Do not write 2,000 words of SERP recap and hope an engine finds a needle.
Do not stuff "Perplexity" and "AI Overview" into every sentence. Name them where the tactic differs.
Do not promise "get cited." Improve odds. Observe. The GEO guide covers measurement.
Do not humanize until Experience (your method, your number, your limit) is on the page.

Bottom line

Answer engines quote blocks, not vibes. Write a definition, a step list, and a dated proof. Human Writes is the rhythm pass after those three exist.

Paste the stiff setup on Human Writes when the boxes are locked.