How to get cited by ChatGPT, Perplexity, and Google AI Overview

The problem

You have a website. You have content. It ranks well on Google. But when someone asks ChatGPT or Perplexity something about your industry, your brand doesn't appear in the answer.

This problem affects most websites in 2026. Generative search engines read content from the web and synthesize it into answers. But they don't cite everything: they select the most suitable sources. If your content doesn't have the right characteristics, it gets ignored even if it's well-positioned on Google.

If you're not clear on what GEO is, read What is GEO: a guide to Generative Engine Optimization first. This guide covers the practical steps for getting cited.

How source selection works

ChatGPT, Perplexity, and Google AI Overview use different approaches, but the underlying logic is the same.

ChatGPT (with browsing enabled) runs web searches in real time, reads the resulting pages, and synthesizes a response. It cites sources at the bottom or inline. It tends to prefer content with specific data and clear definitions.

Perplexity is built as an answer engine: for every response it cites sources explicitly, with numbered links. It uses a mix of web search and its own index. It favors structured, up-to-date content with an identified author.

Google AI Overview appears in the SERP as a generative box above organic results. It draws primarily from content that already ranks in the top positions. Google ranking signals (E-E-A-T, backlinks, structure) directly influence selection.

The common denominator is that all three favor content that answers in a direct, structured, and verifiable way. Understanding the differences between GEO and traditional SEO can help you adapt your strategy accordingly.

Pattern 1: answer the question in the first 200 characters

The most effective pattern is also the simplest. When a user asks a generative engine a question, the system looks for pages that answer that question. If your page answers in the opening paragraphs with a clear sentence, it's more likely to be selected.

A concrete example.

Question: "What is bounce rate?"

Content that doesn't get cited:

"The world of digital marketing is constantly evolving. Among the metrics that professionals monitor daily, there's one that is often underestimated..."

Content that gets cited:

"Bounce rate is the percentage of visitors who leave a site after viewing only one page. It is calculated by dividing single-page sessions by total sessions."

The second version works because the generative engine can extract a precise definition and attribute it to a source. The first version contains no extractable information in its first 200 characters.

Figuring out which structure works for a given keyword isn't always obvious. The most solid approach is to analyze the content already ranking in the top 10 and identify the common patterns: how they open sections, what kind of definitions they use, how direct they are.

Pattern 2: verifiable data

Generative engines have a structural problem: they can produce inaccurate information (so-called "hallucinations"). To reduce this risk, they prefer to cite sources that contain specific, verifiable data.

What counts as verifiable data:

• Numbers with context: "67% of mobile users abandon a site that takes more than 3 seconds to load (Google, 2023)"
• Specific dates: "As of March 2024, Google AI Overview is available in 120 countries"
• Names and references: "According to the Princeton and Georgia Tech study on GEO published in 2024"
• Quantified comparisons: "The conversion rate went from 2.1% to 3.8% after restructuring the content"

What doesn't count: vague percentages without a source, generic claims ("many studies show that..."), rounded numbers without context.

Pattern 3: heading structure as a table of contents

Generative engine crawlers parse the HTML structure of content. Headings (H2, H3) work as a table of contents: they tell the system which topics the page covers and where to find each piece of information.

An effective structure has these characteristics:

• Each H2 corresponds to a specific question or topic
• The first paragraph under each H2 contains the direct answer
• H3s under an H2 expand on specific sub-topics
• The hierarchy is consistent: H2 for macro topics, H3 for details

Structure that works:

H2: What is bounce rate
  -> Definition in 2 lines
  -> How it's calculated
H2: How to reduce bounce rate
  H3: Loading speed
  H3: Mobile experience
  H3: Above-the-fold content
H2: Average bounce rate by industry
  -> Table with data

Structure that doesn't work:

H2: Introduction
H2: Our approach
H2: Deep dive
H2: Conclusions

The second structure contains no information in the headings. The generative engine doesn't know what it will find in each section.

Pattern 4: tables and lists

Tables and ordered lists are the formats most easily parsed by generative engines. When information can be presented in a table, the table is almost always preferable to prose for citability.

Example. If you're comparing three options, a table with columns and rows is more citable than three paragraphs of text. If you're listing steps in a process, a numbered list is more citable than a paragraph describing the same process in prose.

This doesn't mean all content should be in tables and lists. It means that whenever you have comparable data or sequential processes, that format gives you a citability advantage.

Pattern 5: E-E-A-T signals

E-E-A-T stands for Experience, Expertise, Authoritativeness, Trustworthiness. Google has used these signals for ranking for years, and generative engines also use them for source selection.

The E-E-A-T signals most relevant for GEO:

• Identified author, with bio and visible credentials on the page
• Domain with history and authority in the sector (backlinks, mentions, domain age)
• Content with cited external sources (studies, reports, official data)
• Visible publication date and last-updated date
• Direct experience: the content shows the author has done the thing they're writing about, not just read about it

Pattern 6: freshness and updates

Generative engines prefer recent content. An article updated in March 2026 is preferred over an identical one dated 2023.

In practice:

• Always show the publication date and last-updated date
• Update data when it changes: if you cite a 2023 statistic and a 2025 version exists, update it
• Remove references to past dates that make the content feel dated ("next year" written in 2023)
• If the content hasn't changed but is still valid, adding "Updated [date]: confirmed" signals freshness

Differences between the three engines

Operational checklist

Before publishing content, check these points:

• Does the first paragraph answer the main question directly?
• Are there at least 3 verifiable data points with sources in the content?
• Do the H2s contain the keyword or the question they answer?
• Are there tables or lists for comparable data?
• Is the author identified with name and credentials?
• Is the publication date visible?
• Is the content up to date with 2025 or 2026 data?
• Does each section start with the most important information, not a preamble?

If the answer to all of these is yes, the content has the characteristics to be cited by generative engines. It's not a guarantee, but it's the necessary starting point.

Measuring results

There's no single tool yet that measures AI citations the way Google Search Console measures ranking. But you can verify manually:

• Search your main keywords on ChatGPT, Perplexity, and Google (with AI Overview active). Does your content appear among the sources?
• Monitor referral traffic from chat.openai.com and perplexity.ai in Google Analytics
• Compare your content with the sources that actually get cited: what do they have that yours doesn't?

Identifying content gaps between your pages and those that get cited is one of the most practical ways to close the distance.

GEO and AI optimization is a new discipline. Measurement tools are still developing. But the patterns that work are already clear, and applying them now gives you an advantage over those who wait.

If you want to start from data rather than intuition, Verbalist analyzes the SERP for your keyword, extracts the patterns of content that gets cited, and generates content structured on those patterns. You can try it on a real case.