---
title: "ChatGPT SEO Ranking Factors and Signals (2026)"
url: https://encited.com/blog/chatgpt-ranking-factors
description: "ChatGPT SEO ranking factors from our research: authority, Google rank, title and snippet, structure, freshness and mentions, with a page scorecard."
lang: en
---

Image: ChatGPT SEO Ranking Factors: The Signals That Decide What It Cites (2026) (https://encited.com/blog/thumbnails/chatgpt-ranking-factors.webp)

# ChatGPT SEO Ranking Factors: The Signals That Decide What It Cites (2026)

Nov 12, 2026 November 12, 2026 · by Aki from Encited

ChatGPT SEO ranking factors from our research: authority, Google rank, title and snippet, structure, freshness and mentions, with a page scorecard.

Your buyers ask ChatGPT a question your page answers, and you want that page cited. OpenAI publishes no list of ChatGPT ranking factors, so this list comes from our research. One part compares the pages ChatGPT cited with the pages it passed over. The other reads what ChatGPT's servers send while it answers. Each ChatGPT ranking signal below has both: what gets cited more, and why.

**TL;DR.** The ChatGPT ranking factors with the strongest correlations are referring domains, Google rankings, fresh content, and mentions on Reddit, review sites and "best of" lists. Our research shows why. ChatGPT answers from a title, a URL and about 200 characters per page, and it opens the page itself about once in 80 results. Free accounts draw mostly on OpenAI's own index and paid thinking mode draws mostly on Google's results. So allow OAI-SearchBot, rank on Google's first page, state the answer right after your H1, and get named on the pages ChatGPT already reads.

## How ChatGPT search works vs traditional search engines

Google scores a page with its own ranking systems. Research on Google's code counts 305 sets of ranking signals, 48 of them for a web page: rankers, expertise, topic authority, freshness and demotions.

ChatGPT search runs a shorter chain:

1. It rewrites the prompt into one or more searches, called fan-out queries.
2. It sends them to its search sources. Each result comes back as a title, a URL and a snippet.
3. It merges the result lists with Reciprocal Rank Fusion, which favors a page that shows up for several of the searches. A ranking model then scores the candidates against the question.
4. It writes the answer from what is left and cites some of it.

### The account decides which index answers

OpenAI runs its own search index next to the Google results it buys from scraping providers. The same prompts, sent through each ChatGPT mode in August 2026:

| ChatGPT mode | URLs pulled per conversation | From OpenAI's own index | From Google's results |
| --- | --- | --- | --- |
| Free account, Think | 35.3 | 74.7% | 25.3%, most of it news |
| Paid account, thinking | 35.1 | 24.7% | 75.3% |

A free account in instant mode pulled 15.1 URLs. In our research, OpenAI's own index supplied nearly all of them for product, local and evergreen questions, and about half for questions whose answer changes within the day.

Most prompt tracker tools ask their questions from the signed-out version of ChatGPT, which varies vastly from what people see when they are signed in, and the difference is even bigger for people with paid accounts.

## Core ChatGPT SEO ranking factors and signals

| Ranking factor | What gets cited more | What ChatGPT does |
| --- | --- | --- |
| Referring domains and domain trust | Sites above 32,000 referring domains are 3.5 times more likely to be cited | Authority reaches ChatGPT through Google's results and the domains it searches by name |
| Google rankings | Pages at an average position of 1 to 45 get about 60% more citations | One in three URLs ChatGPT used is on Google's first page |
| Title and opening text | Keyword-heavy titles get half the citations of plain ones | ChatGPT sees the title and 202 characters of the page |
| Content length and depth | 5.1 citations above 2,900 words, 3.2 below 800 | ChatGPT opens one page in 80. Length helps a page rank for more fan-out queries |
| Section structure | Sections of 120 to 180 words get 70% more citations | A direct answer at the top of a section scores above 99% on ranking models |
| Freshness | 6.0 citations when updated in the last three months, 3.6 otherwise | The stored snippet changes only when OAI-SearchBot crawls again |
| Page speed | 6.7 citations with FCP under 0.4 seconds, 2.1 over 1.13 | ChatGPT reads stored copies, and rejects pages over 4 MB |
| Schema and llms.txt | No lift from either | JSON-LD is removed before an opened page reaches the model |
| Reddit and Quora mentions | About 4 times the chance of a citation | Reddit threads are pulled often and rarely shown |
| Review platforms | 3 times the chance of a citation | Local answers are built from Yelp, TripAdvisor and Google Maps |
| "Best of" lists | Listicles are 18.8% of citations | Rank 1 in a cited list lifts brand mentions by up to 16.5 points |

Crawler access and the off-site factors have their own sections further down.

### 1. Referring domains, domain trust and traffic

Referring domains were the strongest factor in our research. Sites with up to 2,500 referring domains averaged 1.6 to 1.8 citations, and sites above 350,000 averaged 8.4. Domains with a trust score above 90 got almost four times the citations of those below 43. Traffic mattered only past 190,000 monthly visitors. A `.gov` or `.edu` domain gave no lift: those sites averaged 3.2 citations against 4 for the rest.

The same shape shows in B2B SaaS. 42.6% of ChatGPT's citations went to domains rated DR 80 or higher. Pages with zero traffic still took 11% of citations.

ChatGPT's results carry no link count. They carry a title, a URL and a snippet, and they come from Google, which ranks on links, or from OpenAI's own index. The second route for authority is the `site:` operator. In August 2026 it was in 58.1% of the searches paid thinking mode ran at high effort, and 84% of the domains those searches target are brand domains. A domain ChatGPT already names gets searched directly.

Earn links and coverage from sites in your field. They move you up in Google, and they put your name on the pages ChatGPT reads.

### 2. Google rankings

Research shows pages at an average Google position of 1 to 45 got 5 citations, and pages at 64 to 75 got 3.1. In B2B SaaS, 58% of the pages in Google's top 10 were also cited by ChatGPT.

ChatGPT doesn't run Google's ranking algorithm, and it still depends on Google's results. Our research replayed ChatGPT's own searches on Google and on Bing. One in three of the URLs ChatGPT used was on Google's first page. One in twenty was on Bing's. When a site lost its Google rankings in early 2026, ChatGPT cut its citations at the same time.

A first-page Google position carries the paid mode. A place in OpenAI's own index carries the free ones, and that index is filled by OAI-SearchBot.

### 3. Title, H1 and the opening 200 characters

Our research scored how closely each title and URL matched the target keyword. The closest matches averaged 2.8 citations for titles and 2.7 for URLs. Titles and URLs that named the broader topic averaged 5.9 and 6.4.

ChatGPT judges most pages on three fields: the title, the URL and a snippet. The snippet in OpenAI's own index is cut at 202 characters and comes from the start of the page body. Our research compared 463 cited pages that have an H1 with their stored snippets, and 387 snippets contained the H1. That is 83.6%.

The median H1 is 51 characters, which leaves about 146 characters for your content. The snippet is stored when OAI-SearchBot crawls the page and is the same for every question.

- OpenAI's own index ignores the meta description. Google's results carry it about one time in three.
- OpenAI's own index passes the whole title. Google's results cut it at 60 characters.
- One page in seven had no H1. Its snippet then starts at whichever subheading the template puts first.

Anything between the top of the body and the first sentence uses part of the 202 characters:

| What uses the start of the snippet | Pages affected | Characters used |
| --- | --- | --- |
| Section label or category | 29% | 18 |
| Publication or update date | 11% | 25 |
| Alt text of the first image under the title | 9% | 50 |
| Site or brand name | 7% | 44 |
| Breadcrumb | 7% | 23 |
| Author byline | 3% | 45 |

Write a title that names the topic in plain words, mark up one H1, and put the sentence that states the page's answer right after it.

The ChatGPT crawler simulator (https://encited.com/free-tools/crawler-simulator/chatgpt-crawler-simulator) fetches any URL as OAI-SearchBot and shows the HTML and text it gets, so you can read what follows your H1 the way the crawler does.

### 4. Content length and depth

In our research, articles above 2,900 words averaged 5.1 citations and articles under 800 averaged 3.2. Pages with 19 or more statistics averaged 5.4 against 2.8, and pages with expert quotes 4.1 against 2.4.

ChatGPT reads very little of that length. It opens a page itself in about one result out of 80, and works from the snippet for the rest. Length helps earlier in the chain: a page that covers more sub-questions ranks for more fan-out queries, and Reciprocal Rank Fusion lifts a page that ranks for several.

Depth pays off when the page does get opened. ChatGPT cites a page it opened 74% of the time, against 7% for a page that is only in its results. The pages it opened were primary sources deep inside a site: documentation, pricing pages, regulatory pages and newsrooms.

Cover each sub-question of the topic on the page, and put your own numbers on it.

### 5. Section structure, question headings and FAQs

Research shows pages with sections of 120 to 180 words between headings averaged 4.6 citations, and pages with sections under 50 words averaged 2.7. FAQ sections and question-style headings showed no lift on their own: 3.8 citations with an FAQ section against 4.1 without, and 3.4 for question headings against 4.3 for plain ones.

Named entities count too. Cited pages carried about 10 per 1,000 words against 8 for pages that ranked and weren't cited, and pages naming six or more brands averaged 2.13 citations against 1.21 for pages naming none.

Our passage test shows what the ranking model rewards inside a section. It scored two passages against the search "what are the best aeo tools" on 12 open ranking models. A product description scored near zero on most of them. A one-sentence shortlist that named the tools scored above 99%.

Give each section one sentence that answers its question, in the first two lines. The shape of that answer follows the question:

| The question | The answer shape |
| --- | --- |
| What is X? | A short definition with the category and the facts that set it apart |
| Best X for Y | A shortlist with the criteria for each pick |
| X vs Y | The same criteria applied to both, with the differences stated |
| How to do X | Ordered steps with the expected result |
| Price, date or status | The exact value and the date it took effect |

There is no public evidence of which ranking model ChatGPT runs, so compare two versions of your own passage and keep the one that scores higher on several models.

### 6. Topic coverage and the searches behind the prompt

Our research found ChatGPT rewrites a buyer's question into a keyword string 29% shorter than the prompt, and the words it added most were "practices", "best", "official" and "2026". Research on 5 million fan-out queries agrees: ChatGPT adds "best" to 24.3% of its searches on advice questions, and the current year to 5.44% of prompts.

A page is a candidate only when it ranks for one of those rewritten searches. A cluster of pages on one topic gives ChatGPT a page for each of them.

Write a page for each fan-out query that keeps coming up, and keep the year in the ones whose answer changes yearly.

### 7. Freshness

Content updated in the last three months averaged 6.0 citations in our research, against 3.6 for older pages. Brand-new pages did little better than old ones, so the update is what counts. In B2B SaaS the median cited page was 3.9 months old.

Two stored copies of your page age separately:

- The index snippet changes only when OAI-SearchBot visits again. 13% of the snippets in our research were more than a month behind the crawl.
- The full-page copy is reused for about 30 minutes before ChatGPT fetches the page again, and old copies were still served after 90 days.

Words like "today" or "right now" in a prompt changed nothing measurable. ChatGPT runs a live search when the topic itself changes within the day, like scores, prices and news.

Update the page when its facts change, show the date, and link the page from one that OpenAI's crawlers already visit so the new version gets fetched.

### 8. Page speed and Core Web Vitals

Pages with a First Contentful Paint under 0.4 seconds averaged 6.7 citations in our research, and pages over 1.13 seconds averaged 2.1. Interaction to Next Paint ran the other way: the fastest pages got the fewest citations.

ChatGPT answers from stored copies of a page, so speed counts at the moment a crawler fetches it. INP measures the response to a click, and OpenAI's crawlers don't run JavaScript or click. One hard limit is measured: a page over 4 MB is rejected whole.

Serve the full HTML fast enough that a crawler gets it on the first request.

### 9. Schema markup and llms.txt

Neither showed a lift. Pages with FAQ schema averaged 3.6 citations against 4.2 without it, and an llms.txt file made no difference in our research.

When ChatGPT opens a page, the JSON-LD is removed before the text reaches the model. That says nothing about what the index does with it. Keep structured data for Google and Bing, where it is documented.

## Technical access: OAI-SearchBot and GPTBot

OpenAI runs three crawlers, and each one fills a different store:

| Crawler | What it fills | What blocking it does |
| --- | --- | --- |
| `OAI-SearchBot` | The search index and its stored snippets | Your pages leave ChatGPT's search answers |
| `ChatGPT-User` | The full-page copy ChatGPT reads | Little: people start these requests, and OpenAI says robots.txt rules may not apply |
| `GPTBot` | Training data. In our research it filled neither the index nor the page copy | Your content is marked as not for training. Search stays as it is |

Allow the search crawler, and block the training crawler only if you want out of training:

txt

Copy Download

User-agent: OAI-SearchBot

Allow: /

User-agent: GPTBot

Disallow: /

Sitemap: https://www.example.com/sitemap.xml

Then check that your firewall lets the search crawler's user agent through:

bash

Copy Download

curl -s -o /dev/null -w "%{http_code}\n" \

-A "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36; compatible; OAI-SearchBot/1.4; +https://openai.com/searchbot" \

https://www.example.com/pricing

A 200 means a rule on the user agent didn't stop it. A 403 or a challenge page means one did. OpenAI's real requests come from its published IP ranges, so allowlist those too.

Four more facts from the logs:

- **OpenAI's crawlers don't run JavaScript.** Content has to be in the HTML your server sends. The sources are in do AI crawlers render JavaScript? (https://encited.com/blog/do-ai-crawlers-render-javascript) If you use Encited for pre-rendering and edge caching, this is already taken care of: crawlers get lean HTML and AI-optimized Markdown with your content in it.
- **Discovery runs on links.** In our research, GPTBot downloaded a test sitemap every day and fetched none of the 8 URLs listed only there. It fetched all 8 pages that had a link pointing to them.
- **`noindex` doesn't keep a page out.** Pages with a `noindex` tag still ended up in OpenAI's page copies.
- **Pages over 4 MB are rejected whole.**

The full setup, from robots.txt to crawl logs, is in SEO for ChatGPT (https://encited.com/blog/how-to-get-pages-indexed-by-openai-chatgpt).

## Entity authority and third-party validation

Our research tracked 95 brands across 278 answers. 66% of the times ChatGPT recommended a brand, it cited no page from that brand's site. The recommendation came from what other pages said.

### Brands ChatGPT already knows

In instant mode, 85% of ChatGPT's citations pointed to URLs that were in none of its search results. Part of those come from maps and product cards, and the rest look like brand home pages the model wrote from memory. Wikipedia counts twice here: it is the fourth most-cited domain in AI answers, and it is in the training data.

### Reddit and Quora mentions

In our research, domains with the most Quora mentions averaged 7.0 citations and domains with almost none averaged 1.7. Reddit showed the same range. Reddit is also the most-cited domain across AI engines in research on 30 million sources.

Inside ChatGPT, Reddit threads are pulled often and rarely shown as a citation: they were 2.1% of citations in B2B SaaS. A thread shapes what ChatGPT says about you more than it sends clicks.

### Review platforms

Domains with profiles on Trustpilot, G2, Capterra, Sitejabber or Yelp averaged 4.6 to 6.3 citations in our research, against 1.8 for domains with none. Review platform pages themselves were only 1.6% of citations in B2B SaaS, and review count was one of the four drivers of brand recommendations.

### "Best of" lists

Listicles were the most-cited page type in B2B SaaS at 18.8% of citations, and presence in third-party listicles was the top driver of recommendations. In our research, 52% of the listicles ChatGPT retrieves are cited more than twice per answer, and rank 1 in a cited third-party listicle lifts the chance your brand is mentioned by up to 16.5 points.

### Local and product answers

Local and product answers come from separate systems. Local results are built from Yelp, TripAdvisor and Google Maps listings. Product cards come from merchant feeds and share no URLs with ChatGPT's web results, so a good web ranking doesn't put a product in the carousel.

To see which pages ChatGPT cites for your own prompts, connect Encited MCP (https://encited.com/seo-mcp#connect) to Claude or ChatGPT and ask for the domains AI cites and the fan-out queries behind each answer.

## ChatGPT ranking factors scorecard

Score one page at a time against the search you want it cited for. Generative engine optimization (GEO) for ChatGPT is the work of turning each fail into a pass.

| Check | Pass when |
| --- | --- |
| Crawler access | `robots.txt` and the firewall allow OAI-SearchBot |
| Text without JavaScript | A sentence from the page is in the server HTML |
| H1 | The page has exactly one |
| Text before the H1 | 20 characters or fewer |
| First 150 characters after the H1 | They state the page's answer |
| Title | It names the topic in plain words and reads complete in 60 characters |
| Direct answer | The first two lines of each section answer its heading |
| Answer shape | A list, table or steps that match the question |
| Section length | Most sections run 120 to 180 words |
| Updated date | The page changed in the last three months and shows the date |
| Internal link | A crawled page links to this one |
| Page size | Under 4 MB |
| Google ranking | First page for the fan-out queries |
| Fan-out coverage | A page of yours answers each recurring fan-out query |
| Listicles | The cited "best of" lists name you, near the top |
| Reddit and reviews | Threads and review profiles describe you as you are now |
| Reference facts | Wikipedia and directories state current facts |
| Local listings | Yelp, TripAdvisor and Google Business Profile are complete |
| Product feed | Your merchant feed is submitted, for stores |

You can check most of this scorecard on your own site today. SEO Spider (https://encited.com/site-audit) audits every page for the on-page rows: missing or extra H1s, title length, meta descriptions, thin content, orphan pages and oversized HTML. If you haven't run it in a while, this is a good time to re-run the audit. For the Google ranking row, the Check SERP position content workflow shows which of your pages rank for a search and where. For your brand's presence, set up prompts and track whether ChatGPT names you. All of these are available on all paid plans.

If you use Encited for pre-rendering and edge caching, the low-content rows are already taken care of. Crawlers get lean HTML and AI-optimized Markdown with your full content in them.

## ChatGPT ranking factors: FAQ

**What are the core ChatGPT ranking factors and signals?** Referring domains and Google rankings show the strongest correlations with ChatGPT citations, followed by fresh content and mentions on Reddit, review sites and "best of" lists. On the page, the signals are crawler access for OAI-SearchBot, a title and opening 200 characters that state the answer, and a direct answer at the top of each section.

**What is the difference between OAI-SearchBot and GPTBot in robots.txt?** OAI-SearchBot fills the index behind ChatGPT search, and blocking it removes your pages from search answers. GPTBot collects training data, and blocking it leaves search as it is.

**Does ChatGPT rely on Google's ranking algorithms for search citations?** It doesn't run them, and it uses their output. Paid thinking mode took 75.3% of its URLs from Google's results in August 2026, and one in three URLs ChatGPT used was on Google's first page. Free accounts draw mostly on OpenAI's own index.

**How does ChatGPT select which brands or sources to recommend in answers?** It names brands the model already knows, and it reads the pages its searches return: the brand's own pages, editorial reviews and "best of" lists. 66% of brand recommendations in B2B SaaS came with no citation of the brand's own site.

**Do backlinks and unlinked brand mentions impact AI search visibility?** Both correlate with citations. Sites above 32,000 referring domains were 3.5 times more likely to be cited, and domains with heavy Reddit and Quora mentions about 4 times. Links work through Google rankings and crawler discovery, and mentions work through the pages ChatGPT reads.

**How can you format content so generative AI models can easily parse and cite it?** One H1, the answer in the first 150 characters after it, sections of 120 to 180 words, and each section opening with a sentence that answers its heading. Use a list for a shortlist, a table for a comparison and numbered steps for a process, all in HTML that loads without JavaScript.

**Does ChatGPT favor big brands?** Large sites are cited more: past 190,000 monthly visitors, citations rise with traffic. Below that line traffic made no difference in our research, and pages with zero traffic still took 11% of citations.

**Is ChatGPT SEO the same as Google SEO?** Most of it overlaps, since Google's results feed ChatGPT. The differences are OpenAI's own index, the 202-character snippet, and the weight of third-party mentions.

**How do you rank higher in ChatGPT?** Pick the prompts your buyers ask, collect the fan-out queries and cited pages behind them, and work through the scorecard for each page. The month-by-month process is in how to rank in ChatGPT and other LLMs (https://encited.com/blog/how-to-rank-in-llms).

Image: Google (https://encited.com/crawler-logos/google-search.svg)

Image: Gemini (https://encited.com/crawler-logos/gemini.png)

Image: ChatGPT (https://encited.com/crawler-logos/chatgpt.svg)

Image: Claude (https://encited.com/crawler-logos/claude.svg)

Image: Grok (https://encited.com/crawler-logos/xai.svg)

Image: Meta (https://encited.com/crawler-logos/meta.svg)

Image: Perplexity (https://encited.com/crawler-logos/perplexity.svg)

## Get discovered anywhere search happens

Readable, citable, outranking pages.

Talk to sales: https://encited.com/contact-sales

Image: Avatar (https://replymaven.com/api/uploads/Su6Q0nK63iZW0ZFJBRNWpmRuppVDLuab/98df5262-f992-45fa-a750-00a6c919baf9.png)

How can we help?

Get instant answers to your questions or leave a message for an engineer will reach out

Ask AI about Encited

See our docs (https://lovablehtml.com/docs)

Contact support

Leave a message

We'll get back to you soon

Ask AI about Encited

Product questions, troubleshooting and

Thinking

Image: Preview

Drop an image to attach

Powered by ReplyMaven (https://replymaven.com/)

## Structured data

```json
[
  {
    "@context": "https://schema.org",
    "@type": "BreadcrumbList",
    "itemListElement": [
      {
        "@type": "ListItem",
        "position": 1,
        "name": "Home",
        "item": "https://encited.com/"
      },
      {
        "@type": "ListItem",
        "position": 2,
        "name": "Blog",
        "item": "https://encited.com/blog"
      },
      {
        "@type": "ListItem",
        "position": 3,
        "name": "ChatGPT SEO Ranking Factors: The Signals That Decide What It Cites (2026)",
        "item": "https://encited.com/blog/chatgpt-ranking-factors"
      }
    ]
  },
  {
    "@context": "https://schema.org",
    "@type": "Article",
    "headline": "ChatGPT SEO Ranking Factors: The Signals That Decide What It Cites (2026)",
    "description": "ChatGPT SEO ranking factors from our research: authority, Google rank, title and snippet, structure, freshness and mentions, with a page scorecard.",
    "image": "https://encited.com/blog/thumbnails/chatgpt-ranking-factors.webp",
    "datePublished": "2026-11-12",
    "dateModified": "2026-11-12",
    "author": {
      "@type": "Person",
      "name": "Aki from Encited"
    },
    "publisher": {
      "@type": "Organization",
      "name": "Encited.com",
      "alternateName": "LovableHTML.com",
      "logo": {
        "@type": "ImageObject",
        "url": "https://encited.com/brand/logo.svg"
      }
    },
    "mainEntityOfPage": {
      "@type": "WebPage",
      "@id": "https://encited.com/blog/chatgpt-ranking-factors"
    }
  }
]
```