How to get cited by ChatGPT, the 2026 playbook
ChatGPT is the highest-traffic surface in AI search. As of Q1 2026 it processes roughly 3.8 billion weekly conversations (OpenAI quarterly disclosure, openai.com), and a growing share of those conversations include the search tool, which means the model is selecting citations from a real-time index every time it answers a factual question. If your brand is not in that citation set, you are invisible to the largest single channel in AI-mediated discovery.
This guide explains exactly how ChatGPT selects citations, what bots crawl your site on its behalf, and the 8 steps that move a brand from invisible to consistently cited.
How ChatGPT picks citations
ChatGPT uses three retrieval pathways depending on the conversation context.
**Pathway 1, training data citation.** When the model answers from pretrained weights without invoking search, it can still attribute information to a source if the source was prominent in the training set. This pathway favors domains with large historical footprints (Wikipedia, Reuters, established industry publications).
**Pathway 2, live Bing retrieval.** When the search tool is triggered (either by a user toggle or by ChatGPT's own routing logic on time-sensitive queries), the model runs a Bing search, retrieves the top results, and constructs an answer with inline citations. This pathway is the one most brands can actually influence in the short term, because Bing rankings are knowable and movable.
**Pathway 3, ChatGPT-User agent fetches.** When ChatGPT itself decides to read a specific URL (often from a user-pasted link or a Bing top result), it sends the ChatGPT-User agent to fetch that page and incorporate its content into the answer. Blocking this agent makes your page unciteable on a per-conversation basis.
The practical implication: ranking on Bing is the floor for ChatGPT citation eligibility. You can be cited from pathway 1 (training data) without ranking, but for any query that triggers search, your Bing position determines whether ChatGPT ever sees your content.
ChatGPT's bot user agents
Four agents matter for ChatGPT-related crawling and retrieval.
| Bot | Purpose | Should you allow? |
|---|---|---|
| GPTBot | Trains future OpenAI models on your content | Yes if you want to be in training data |
| OAI-SearchBot | Indexes pages for ChatGPT's search feature | Yes, required for live citation |
| ChatGPT-User | Fetches a specific URL during a conversation | Yes, required for direct retrieval |
| Bingbot | Crawls for Bing index (ChatGPT's search backbone) | Yes, this is the upstream dependency |
Confirm allowlist in your robots.txt. The minimum allow-block for ChatGPT visibility is:
```
User-agent: GPTBot
Allow: /
User-agent: OAI-SearchBot
Allow: /
User-agent: ChatGPT-User
Allow: /
User-agent: Bingbot
Allow: /
```
OpenAI publishes the IP ranges and user-agent strings at platform.openai.com/docs/bots. Verify your firewall rules do not silently block these ranges.
The Bing search backbone
Bing is the most under-rated SEO surface of 2026. Roughly 35 to 40% of ChatGPT search-tool invocations resolve to a Bing top-10 result that becomes a citation. Microsoft Copilot uses Bing directly. Yahoo Search uses Bing. DuckDuckGo uses Bing.
What ranks on Bing differs from what ranks on Google in three measurable ways:
1. Bing weights exact-match keyword usage more heavily than Google's semantic intent matching.
2. Bing favors older domains with stable backlink graphs (less churn-friendly than Google).
3. Bing's image and video integration into SERPs is broader; mixed-media pages earn more SERP real estate.
If your Bing rankings lag your Google rankings by 30+ positions on commercial queries, you are bleeding ChatGPT citations every day. Bing Webmaster Tools is free; submit your sitemap and run the URL Inspection tool against your top 20 commercial pages as a baseline diagnostic.
What ChatGPT cites most
OnyxRank ran 4,200 ChatGPT search-tool scans across diverse buyer-intent queries in April and May 2026. The domain types that appeared in citation lists ranked as follows.
| Rank | Type | Share of citations |
|---|---|---|
| 1 | Reddit threads (UGC) | 21% |
| 2 | YouTube videos and transcripts | 13% |
| 3 | Industry publications (Search Engine Land, TechCrunch, etc) | 12% |
| 4 | Long-form blog posts on incumbent SEO/marketing domains | 11% |
| 5 | Wikipedia | 8% |
| 6 | G2, Capterra, TrustRadius review pages | 7% |
| 7 | Brand owned long-form content | 6% |
| 8 | Government and academic .gov / .edu | 4% |
| Other | Forums, niche blogs, social | 18% |
The lesson is unambiguous: ChatGPT prefers third-party validation surfaces (UGC, reviews, industry pubs) over brand-owned pages by a roughly 3 to 1 margin. If your entire AI SEO strategy is "publish more blog posts," you are working against the citation distribution.
The 8-step ChatGPT action plan
1. **Audit your Bing rankings.** Pull Bing Webmaster Tools data for your top 50 commercial keywords. Any keyword ranking position 20 or worse on Bing is a ChatGPT citation you cannot win.
2. **Allowlist all four bot agents.** GPTBot, OAI-SearchBot, ChatGPT-User, Bingbot. Verify with `curl -A "GPTBot" https://yourdomain.com` and confirm a 200 response, not a 403 from CDN bot protection.
3. **Ship a /llms.txt and /llms-full.txt.** The /llms.txt file is the AI-era equivalent of robots.txt for discovery. /llms-full.txt is the long-form knowledge base. Both are now supported by OpenAI's crawler logic; pages discovered through these files have a higher index priority than pages found via random link traversal.
4. **Earn Reddit citations.** Reddit accounts for 21% of all ChatGPT citations on commercial queries. Identify the 5 subreddits where your buyers ask questions, contribute meaningfully (not promotionally) over a 30-day period, and earn the right to reference your brand in context. This is not optional in 2026.
5. **Get listed on the canonical review platforms.** G2, Capterra, TrustRadius if you are B2B SaaS. Trustpilot if you are DTC. Listings on these surfaces drive 7% of citations on average.
6. **Publish YouTube content.** Even 5 short videos (under 3 minutes) per quarter with accurate transcripts move the citation needle, because ChatGPT pulls transcript content into answers. A YouTube channel with 200 subscribers and good transcripts beats a 2,000-page blog with no video presence for video-eligible queries.
7. **Structure your answer-first content.** Within your top 20 commercial pages, the answer to the implied buyer question should appear in the first 1 to 2 sentences below the H2 that mirrors the question. ChatGPT does not read your introduction; it extracts answer spans.
8. **Run a weekly Pulse scan.** Pick 10 to 20 buyer-intent prompts and run them against ChatGPT with the search tool enabled, weekly. Track which prompts surface your brand, which engine cited which competitor, and how the share shifts over 30 to 60 day windows. A free version of this scan is at https://onyxrank.com/tools/citation-checker.
Common mistakes
The four mistakes we see most often in audits.
**Blocking ChatGPT-User in robots.txt because it "is not a search crawler."** It is not, technically. It is a real-time retrieval agent. Blocking it makes your URL unciteable inside conversations even if you rank on Bing.
**Optimizing only for Google.** Brands that have spent 5 years optimizing for Google often score 60+ on Google citation eligibility and 15 to 25 on Bing. The Bing gap is the ChatGPT gap.
**Treating Reddit as a marketing surface.** Reddit AutoMod and community moderators detect promotional comments fast. The accounts that earn citations are accounts with 6+ months of unrelated comment history before they reference a brand. Plan accordingly or skip the channel.
**Publishing answer-burying long-form.** Pages that open with 300 words of context before the answer are systematically passed over by ChatGPT in favor of pages that lead with the answer. Restructure, do not just publish more.
How to verify you are working
Three measurable checks.
1. **Direct citation count.** Run the same 10 prompts in ChatGPT with the search tool weekly. Record the citation list for each. The number of times your domain appears across the 10 results is your raw citation count. Track week-over-week.
2. **Bing position parity.** Pull your top 20 commercial keyword positions on both Bing and Google. The gap between the two is the share of ChatGPT citations you are leaving on the table.
3. **Bot fetch logs.** Look at your access logs filtered to user-agent containing "GPTBot", "OAI-SearchBot", and "ChatGPT-User". A healthy AI-visible site has all three agents fetching pages at least weekly, and ChatGPT-User specifically fetching pages within minutes of a relevant query being asked.
What to ship next
Once the 8 steps are in motion, the next leverage is engine-specific surface area. Pages structured for ChatGPT (Bing-eligible, answer-first, third-party validated) compound across the rest of the AI search stack: Copilot, Bing organic, Yahoo, and DuckDuckGo all benefit from the same investment.
For a free audit that runs the exact 10-prompt scan described in step 8, visit https://onyxrank.com/tools/citation-checker. For the full ranking strategy across all 4 engines, see https://onyxrank.com/blog/ai-citation-formula-geo-optimization-2026.
Pro Intel subscribers get the full picture - proprietary analysis, keyword opportunities, tactical playbooks, and template downloads every week. $49/mo.
One email per week. Actionable, no fluff.