How to Rank
in ChatGPT.
ChatGPT search doesn’t have its own index. It uses Bing’s. So “ranking in ChatGPT” collapses into two problems most UK businesses don’t separate cleanly.
Most people asking “how do I rank in ChatGPT?” in 2026 are using the wrong mental model. They imagine ChatGPT as a search engine with its own crawl, its own ranking signals, its own SEO playbook. It isn’t. ChatGPT search is a wrapper around Bing’s index, with a separate answer-quality layer applied on top. That changes what you do about it.
Step one: be in Bing’s index.#
In one line Verify in Bing Webmaster Tools, submit the sitemap, implement IndexNow. Twenty minutes. The lever most UK businesses haven’t touched.This is the part most UK businesses skip. Verifying a site in Google Search Console is a habit. Verifying in Bing Webmaster Tools is treated as optional. It isn’t.
Bing’s index powers four surfaces in 2026: Bing.com itself (small market share, ignorable), Microsoft Copilot (embedded in Office, Outlook, and Edge), DuckDuckGo (which licenses Bing results), and ChatGPT search (the one most UK marketers actually care about). If you’re not in Bing’s index, you’re not in any of them.
The setup is twenty minutes:
- Go to bing.com/webmasters and sign in with a Microsoft account.
- Add your site. Verification is a meta tag, a DNS record, or an XML file at the root.
- Submit your sitemap. The same sitemap.xml you submitted to Google Search Console works here.
- Use the URL Inspection tool to request crawling for your highest-priority pages. Same flow as Google Search Console’s “Request Indexing”.
- Implement IndexNow. One API ping notifies Bing the moment you publish a new page or update an existing one. Yandex, Seznam, and Naver get the same ping. The protocol is documented at indexnow.org.
If you’re an F&T client and your site is in our deploy pipeline, IndexNow is wired up automatically. If not, it’s a serverless function and a key file at your domain root. We covered the strategic context in The Multi-Index Problem.
One verification, three surfaces. The structural reason Bing Webmaster Tools is the highest-leverage thirty minutes most UK businesses haven’t spent.
Carry the citation signals ChatGPT weighs.#
In one line Bing eligibility is not Bing citation. Eight signals stack on top: direct-answer format, named authority, fact density, schema, external citations, freshness, domain consistency, HTML semantics.Being in Bing’s index makes you eligible. It doesn’t make you cited. The answer layer ChatGPT applies on top of Bing weighs eight signals, derived from observable patterns in citations across thousands of queries. Listed in rough order of leverage:
- Direct-answer format. The first sentence of the relevant page should answer the query the user typed, not introduce the topic. “What does an AI search audit cost?” pairs with “An AI search audit costs £250 to £1,500 in the UK in 2026, depending on depth.”, not “In an increasingly AI-driven world, businesses are turning to…”.
- Named authority. A real human author with a credential signal on the page. Avatar, name, role, link to a profile or other writing. ChatGPT prefers attributable sources.
- Fact density. Specific, verifiable claims. Numbers, dates, named outcomes. “42 commissioned audits” beats “extensive audit experience”.
- Schema markup. Article schema on editorial, FAQPage on FAQ blocks, BreadcrumbList for hierarchy, Organization with sameAs entries. Not magic. Just structured signals the answer layer can parse cheaply.
- External citation signals. Trade press, comparison sites, Reddit, Wikipedia, directories. ChatGPT triangulates against external mentions when scoring relevance. We covered the structural reason in The Citation Graph Doesn’t Read Your Drafts.
- Freshness. Visible publish and updated dates on the page. Not just in schema, in body copy. The answer layer treats “last updated 2024” as a downgrade signal.
- Domain consistency. One brand name, one tone, used consistently across the site. Variants confuse entity resolution.
- HTML semantics. Proper H1, H2, H3 hierarchy. Lists for lists. Tables for tabular data. Avoid JavaScript-rendered body text on first paint.
| Signal | What it actually means | Lives in |
|---|---|---|
| Direct-answer format | First sentence answers the query. No throat-clearing. No brand intro before the answer. | Body copy, top of page |
| Named authority | A real human author with a credential signal and a link to other writing or a profile. | Author byline + Article schema |
| Fact density | Specific, verifiable claims: numbers, dates, named outcomes, real organisations. | Body copy |
| Schema markup | Article on editorial, FAQPage on FAQ, BreadcrumbList for hierarchy, Organization with sameAs. | JSON-LD in <head> |
| External citation signals | Trade press, comparison sites, Reddit, Wikipedia, sector directories. Triangulation evidence. | Off-site; PR-led |
| Freshness | Visible publish and updated dates in body copy as well as schema. “Updated 2024” is a downgrade signal. | Body copy + schema |
| Domain consistency | One brand name, used consistently. Variants confuse entity resolution. | Site-wide; Organization schema |
| HTML semantics | Proper H1–H3 hierarchy. Lists as lists. Tables as tables. Avoid JS-rendered body text on first paint. | Markup, page template |
What this looks like on a real page.#
In one line Same content, structured differently. The version that ranks has its answer in paragraph one and its schema unbroken. The version that doesn’t, buries both.Take a UK wedding venue page targeting “best wedding venues in Cotswolds 2026”. The version that ranks in ChatGPT has:
- An H1 that includes the query phrase.
- A first paragraph that names three or four specific venues with one-line differentiators.
- Pricing visible as HTML text, not gated behind “contact us”.
- A named author with a venue-industry credential.
- Article schema with datePublished, dateModified, author, and publisher.
- FAQPage schema for the “commonly asked” section.
- Inbound links from a UK trade publication or directory.
The version that doesn’t rank has the same content, structured wrong: a hero image with the H1 baked in as a graphic, a 600-word brand-led intro before the venue list, no schema, no author, contact-form pricing.
What you can’t directly control.#
In one line Training cut-off, Reddit and Wikipedia presence, trade press. Three signals that cannot be gamed quickly. PR matters more in 2026 than it has in a decade.Three signals matter and cannot be gamed quickly:
- OpenAI’s training corpus cut-off. If your domain didn’t exist or was thin during the relevant window, you may be invisible to non-search prompts. The fix is search-mode citations, which use Bing live, not training memory.
- Reddit and Wikipedia presence. ChatGPT cites both heavily. Earning either is slow and editorial.
- Trade press placement. If The Drum, Marketing Week, or your sector’s tier-1 trade publication has never written about you, ChatGPT’s triangulation defaults to whoever has been written about. PR matters more in 2026 than it has in a decade.
How to test progress.#
In one line ChatGPT search mode. Your top queries. Read the citations. Monthly. Visibility gained quietly is also lost quietly.Open ChatGPT. Switch to search mode. Run the queries that matter to your business. Read the citations carefully. The cited domain pattern tells you everything you need to know about which signals are working and which aren’t.
Repeat the test monthly. Drift is real. Bing’s ranking shifts. ChatGPT’s answer layer updates. Competitors publish. Visibility gained quietly is also lost quietly.
“ChatGPT has its own crawler. We’ll wait for our SEO agency to figure it out.”
“ChatGPT runs on Bing. We claim Bing today, fix the eight signals this quarter, and re-test monthly.”
Three patterns that lose ChatGPT citation.#
In one line Composite cases from the F&T audit base. Each one shows a different point at which the two-step problem breaks down silently.The three composites below are not single brands. Each is drawn from multiple Q1 and Q2 2026 audits where ChatGPT cited a competitor on a query the brand should have owned.
Google Search Console verified for eight years. Bing Webmaster Tools never claimed. The firm was technically in Bing’s automatic index, but with stale meta data and no IndexNow notifications when pages updated.
Verification, sitemap submission, and IndexNow took thirty minutes. Within two weeks, ChatGPT cited the firm on its top three commercial queries.
Bing-indexed. Schema present. Author named. Every signal correct except the one that matters most: the direct answer to the buyer’s query was below the fold. The ChatGPT answer layer migrated to a competitor whose category page named three venues in its first paragraph.
Re-ordering paragraphs (no new content) lifted citation from one cell to four within ten days.
Every page-level signal in place: direct answer, fact density, schema, freshness, author. The answer layer still preferred a competitor on most comparison queries. The triangulation evidence was missing; the sector’s tier-1 publications had never mentioned the brand.
Three placements over a quarter (one in The Drum, two in a sector trade publication) shifted the answer layer’s preference on the contested queries.
Three different failures, three different fixes, one shared diagnosis: the brand was treating ChatGPT as one problem when it is two, stacked.
Ranking in ChatGPT is two problems. Solve them in order.#
Get into Bing (verify in Bing Webmaster Tools, submit sitemap, implement IndexNow). Then carry the eight citation signals (direct-answer format, named authority, fact density, schema, external citations, freshness, domain consistency, HTML semantics).
Do the first today. Audit the second this quarter. Re-test monthly. The mistake most UK businesses make is treating “ChatGPT SEO” as one thing. It’s two, in this order.
The citation factors move every quarter.
A monthly intelligence brief on how AI is changing what UK customers find, and what they don’t. No consultancy noise.
Sources & references
- Bing Webmaster Tools, Verification & setup: documentation for site verification, sitemap submission, and the URL Inspection request-indexing flow. bing.com/webmasters/help
- IndexNow, Protocol documentation: the open protocol that notifies Bing, Yandex, Seznam, and Naver of new or updated URLs. Adopted by Cloudflare, Wix, and others as a default. indexnow.org
- OpenAI, Bots & crawlers documentation: canonical reference for GPTBot, OAI-SearchBot, ChatGPT-User and the user-agent split between training, indexing, and live retrieval. platform.openai.com/docs/bots
- Schema.org, Article, FAQPage, BreadcrumbList, Organization: canonical reference for the structured-data types whose presence correlates with citation in our audit base. schema.org
- Microsoft, Copilot in Microsoft 365: product reference for the Office, Outlook, and Edge surfaces that share Bing’s underlying retrieval. learn.microsoft.com
- Friction & Toil audit base, Q1 2026: 42 commissioned Web Intelligence Reports across UK hospitality, professional services, and heritage venues. Source for the “30-minute setup” figure cited in the hero stat strip and the three composites in Section V ½.
- Friction & Toil live re-test, May 2026: 38-brand curl-based crawler-policy and schema scan. Companion evidence for the eight-signal patterns described in Section II.