How Google AI Overviews, Perplexity, and ChatGPT Actually Choose Sources

The three engines pick sources very differently

Google AI Overviews, Perplexity, and ChatGPT all generate answers by pulling from web sources and citing them. They look similar on the surface. Under the hood they are built on completely different logic, and the signals that win a citation on one engine can be invisible on another.

If you are only optimizing for one of them, you are leaving citation share on the table. If you are optimizing for all three at once, you need to understand what each one actually rewards.

This is a working reference for what each engine weights when it decides which sources to cite. I run SEO, AEO, and GEO strategy across 175+ dealerships at Asbury Automotive Group, and the differences below are the ones I see move the needle in real campaigns.

Google AI Overviews: traditional SEO signals still matter, plus a new layer

Google AI Overviews are the AI-generated answers at the top of Google search results. Getting cited inside an Overview delivers the same visibility as ranking #1, without the click.

The engine is built on top of Google's classic search index. That means everything Google already rewards in organic rankings — page speed, backlinks, domain authority, E-E-A-T — is still in play. What is new is that Overviews weight structured extractability much more heavily than traditional rankings do.

What Google AI Overviews prioritize:

• Direct answers in the first 40 to 60 words of a page. The Overview extracts that answer verbatim. Bury the answer under a preamble and you lose the citation.
• Question-formatted H2 headings that mirror real user queries. Overviews treat those headings as extractable answer units.
• FAQPage and Article schema with clear author references. This tells Google's models exactly what to lift and who wrote it.
• Topical clusters — a domain that covers a subject in depth with tightly linked pages beats a one-hit page from a bigger domain.
• Core Web Vitals in the green. Slow pages get skipped even when the content is right.
• [E-E-A-T signals](/answers/what-is-e-e-a-t) — named author with credentials, publish and updated dates, cited sources.

The Overviews algorithm favors domains that combine classic SEO strength with structured, answer-first content. If your site is technically clean and content is structured for extraction, you can compete with domains that have significantly more backlinks than you do.

Perplexity: recency, structure, and speed win over authority

Perplexity is the citation engine most sensitive to freshness. It composes answers from live web sources and shows citations inline. Users click through to sources at much higher rates than they do from Google, which makes every Perplexity citation qualified traffic.

The winning signals here are different from Google:

• Recency is heavily weighted. A page with dateModified set to the last 90 days is treated as fresher than a stale page even if the stale page has more inbound links. Set dateModified in your Article schema and expose an "Updated on" line visible to human readers. Both matter.
• Speed matters more than the SEO industry realizes. Perplexity's crawler has aggressive timeouts. Pages that take more than 2 to 3 seconds to serve initial HTML get skipped entirely.
• Structured data is not optional. Perplexity's extractors read JSON-LD to identify author, publisher, publish date, and headline. Pages without schema rank lower in citation candidacy even when the content is strong.
• Topical depth wins over one-hit content. Ten focused pages on a subject will out-cite one "ultimate guide" that tries to cover everything.
• Named authors with cross-platform presence. Perplexity's ranking models notice when the same author appears across a domain, on LinkedIn, and in third-party mentions.

ChatGPT: entity trust built over time

ChatGPT is the hardest engine to influence in the short term because most of its knowledge comes from its training data cutoff, not live web browsing. Even with browsing turned on, ChatGPT's default posture is to trust what it already knows over what it just fetched.

That means ChatGPT citation is a long-game play. The businesses that get cited by ChatGPT are the ones the model already recognizes as authoritative entities in their space, and that recognition is built over months and years of consistent signals.

What ChatGPT weights:

• Cross-platform entity consistency. Your name, description, services, and credentials should match everywhere the model can see them — your website, LinkedIn, Wikipedia if applicable, Wikidata, industry publications, Google Business Profile.
• Named brand mentions across trusted domains. Backlinks are one form of this. Editorial mentions without links are almost as valuable. Being quoted in industry publications, appearing on podcasts, and being cited in academic or industry reports all build the entity signal ChatGPT uses.
• Depth of published content under your name. ChatGPT's models recognize authors who have consistently published on a topic. A single strong article does less for ChatGPT citation than fifty articles across two years demonstrating sustained expertise.
• Comprehensive coverage of a topic cluster. ChatGPT is more likely to cite a domain that covers a subject exhaustively than one that covers it partially, because comprehensive coverage signals real expertise.
• Schema markup for entity clarity. Person schema with credentials, worksFor, sameAs links to LinkedIn and other platforms, and knowsAbout properties helps ChatGPT build a machine-readable picture of who you are.

When ChatGPT browses live, it favors sources with clear structured data and question-answer format content. That is where AEO overlaps with what ChatGPT rewards. But the deeper play is entity building, which is a discipline closer to GEO than traditional SEO.

Head-to-head: what each engine actually rewards

| Signal | Google AI Overviews | Perplexity | ChatGPT |

|---|---|---|---|

| Direct answer in first paragraph | Critical | Critical | Helpful |

| FAQPage / Article schema | Critical | Critical | Helpful |

| Author schema with credentials | Important | Important | Critical |

| Recency (dateModified) | Moderate | Critical | Low (weight on training) |

| Backlinks / domain authority | Important | Moderate | Important (for entity) |

| Cross-platform entity consistency | Moderate | Moderate | Critical |

| Page speed / Core Web Vitals | Important | Critical | Low |

| Topical depth (cluster of pages) | Critical | Critical | Critical |

| Named brand mentions | Moderate | Moderate | Critical |

| Live browsing / crawl frequency | Continuous | Continuous | On-demand |

The pattern: Google AI Overviews and Perplexity both reward structured, answer-first content that a crawler can extract cleanly today. ChatGPT rewards the same signals but weights the entity story much more heavily, and the entity story takes time to build.

What this means for your content strategy

If you have never optimized for AI citation before, do these four things in order:

1. Front-load direct answers on every page. First 40 to 60 words: what is the topic, and what is the answer. Every page. Every post. Every FAQ entry. This one change alone lifts you on all three engines.
2. Deploy [FAQPage and Article schema](/answers/what-schema-markup-should-i-use) everywhere they fit. With real author references, dateModified, and headline. This is the single biggest structural win for both Google AI Overviews and Perplexity.
3. Build [Person schema](/blog/schema-markup-guide-small-business) for the author. With credentials, worksFor, sameAs links to LinkedIn, and knowsAbout. This is what starts building the ChatGPT entity signal.
4. Pick a topic cluster and go deep. Ten pages on one subject, tightly linked, beats one "ultimate guide" that tries to cover everything. All three engines reward this.

Once those four are in place, the extension work is:

• Speed. Ship server-rendered HTML. Get initial page load under 2 seconds. This is a Perplexity multiplier and does no harm on the other engines.
• Recency signals. Actually update your content and expose dateModified. Perplexity rewards it directly; the other engines treat it as a trust signal.
• Entity building. Get cited in industry publications, quoted on podcasts, mentioned by name in third-party content. This is how you compound ChatGPT citation over time.

The measurement side

The one question I get every time I present this framework: how do I know it is working?

You use AI visibility tracking tools. Profound is the most complete platform I use in enterprise contexts — it tracks citation share across ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews for a defined query list. Otterly.AI and Peec AI are lighter alternatives for small businesses.

For businesses without budget for a paid tool, the low-tech version works: pick 20 high-intent questions your ideal customer would ask, run them monthly against ChatGPT, Perplexity, and Google AI Overviews, and log whether your brand is cited. Track the citation percentage over time. It is manual, but it produces a real trend line.

The engines that don't cite you now are the ones you can still get in front of. The window closes as competitors start doing the work. Right now most industries have almost nobody building for AI citation deliberately.

Start with the audit

Before you change anything, run the four questions on your own pages:

1. Does the first paragraph directly answer the question the page is about?
2. Is there FAQPage or Article schema deployed with a real author and dateModified?
3. Are your H2s formatted as questions users actually search?
4. Does the domain cover this topic in depth, or is this a one-off page?

If the answer to any of those is no, that is the first thing to fix. If the answer to all four is yes, you are already ahead of most of your industry, and the extension work is what compounds you into a durable citation position.

The AI search shift is the biggest change to how people discover businesses in the last fifteen years. The businesses that build for it early lock in citation authority that becomes extremely difficult to displace once the models learn to trust them. The window is still open. That will not last forever.