How to get cited by ChatGPT in 2026

OpenAI reported more than 900 million weekly active users for ChatGPT in March 2026, and said search usage nearly tripled in a year (OpenAI, March 31, 2026). SearchGPT was a prototype launched in July 2024. OpenAI folded it into ChatGPT search in October 2024 and opened search to everyone, with no signup, in February 2025 (OpenAI). The product is no longer a chat toy. It is a research surface, and some buyers now start there. When a user asks ChatGPT for a recommendation, the citation panel is the consideration set. Brands inside it get evaluated. Brands outside it do not.

The shift is structural. The traditional Google research path involved ten blue links, a few clicks, and a longer attention loop where SEO and content marketing did their work. The ChatGPT path is a single conversational turn that produces a synthesized answer with named sources. Citation is the new ranking. If you are not cited, you are not in the conversation, even if you would have ranked first on Google for the same query.

Our AEO audit tool checks observable technical readiness across 13 checks and 66 possible points. It does not predict whether ChatGPT will cite a page. The opportunity is enormous and the bar is currently low.

ChatGPT pulls from two layers. The first is the model’s training data, which is fixed at training time and updates only when OpenAI ships a new base model. The second is the live retrieval layer, ChatGPT search. OpenAI says it uses third-party search providers and content supplied directly by its partners (OpenAI), and ChatGPT uses it when the question needs current information or the user turns search on. Most citation-mode answers happen in the second layer. The first layer matters more for general-knowledge brand recognition.

This is critical because it means two things simultaneously. One: your live web presence determines what retrieval systems can access and evaluate, but no technical checklist guarantees a citation. Two: your historical web presence (long-running brand mentions, established entity coverage) drives the model’s underlying recognition of you as a real entity. The first can be improved directly. The second depends on work accumulated over time.

The source format varies. Links can appear inside the answer, and a Sources button opens the full list. The citation is what the user sees. The model frequently lifts a sentence or short paragraph from each cited source and weaves it into the answer. Pages whose content is hard to lift cleanly (heavy marketing prose, no factual claims, no clean snippets) get cited less, even when the underlying authority is strong.

The criteria differ from Google AI Overviews and from Perplexity. Google says its generative AI features depend on the same core ranking and spam systems as Search (Google Search Central). OpenAI and Perplexity do not publish their source-selection criteria, so treat any claimed weighting for them as a guess. We covered the cross-engine differences in AEO vs SEO explained.

None of these is novel. The AEO playbook is now legible. What is novel is how few sites actually ship all seven. Most stop at two or three and assume that is enough. It is not. No single item guarantees a citation, which is why we treat them as a set.

1. Check OpenAI retrieval access in robots.txt

Start by checking the agents OpenAI uses for search and user-requested retrieval: OAI-SearchBot and ChatGPT-User. If robots.txt blocks them, OpenAI retrieval may not be able to access the page directly. GPTBot is a separate model-training crawler, so blocking it is a training-policy choice rather than an automatic search-access failure. The audit scores retrieval access as an 8-point important check.

The same retrieval check applies to Claude-SearchBot, Claude-User, and PerplexityBot. Google-Extended and training crawlers such as GPTBot are separate controls. The free AEO audit reports whether the named search and user-retrieval agents are blocked on your site.

2. Consider an optional llms.txt file

llms.txt is an optional Markdown discovery file at the root of a domain. Compatible systems may use it to find selected public pages, but adoption is incomplete, Google does not use it, and it cannot guarantee citations. That is why the audit gives it 2 of 66 points.

We wrote the full technical breakdown in llms.txt explained. The short version: H1 with your brand or site name, blockquote describing what the site is, H2 sections for the major content areas (Services, Case studies, Docs, About, Resources), and bulleted links with one-line descriptions. Curate, do not dump. Keep the descriptions factual and use the file as a discovery map rather than a sitemap clone.

3. Organization Schema Sitewide

JSON-LD Organization schema in the head of every page is the entity layer. It tells AI engines that you are a real business with a name, a logo, a website, and verified social profiles. Google says structured data is not required for its generative AI features (Google Search Central), and OpenAI does not say how ChatGPT uses it, so treat schema as a clarity aid, not a switch. The schema lives in a single small JSON-LD block. Most modern frameworks support shipping it sitewide in one line.

For local-focused brands, the right shape is LocalBusiness schema with address, service area, and opening hours. For ecommerce, layer Product, Review, and Offer schemas onto product pages. For editorial content, add Article or BlogPosting schema with named author. The pattern is consistent: one block per page, types appropriate to the page. Free AEO audit shows you which schema types are present.

4. FAQPage Schema Where You Have Visible Q&A

FAQPage can describe visible question and answer pairs, but it is not a ranking factor or a citation guarantee. Add it only where matching Q&A is visible, never as schema-only bait. The audit treats exact FAQ parity as 2 of 66 audit points.

Practical move: add a visible FAQ section when the page has useful buyer questions to answer, then keep the FAQPage markup identical to that visible copy. There is no universal question count, and the markup does not guarantee a citation.

5. Named Author With Bio (E-E-A-T)

OpenAI does not publish how ChatGPT weighs authorship, so nobody outside OpenAI can say named authors earn more citations. We add them anyway, because readers trust an accountable byline. Add a named author byline to every editorial page, link to a person bio page with credentials, and add Person schema where appropriate. The bio page itself should be schema-marked-up and rich enough to be cited as a standalone source. Google encourages accurate bylines and describes E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) as a quality concept, not a specific ranking factor.

The cheap fix: a single <meta name="author"> tag on every page is the floor. Real author byline pages with Person schema and rich content are the ceiling. Most sites we audit fail the floor.

6. Entity Coverage Outside Your Own Site

This is the one most teams underestimate. ChatGPT does not just cite sites; it cites entities. To be a parseable entity, you need coverage on the third-party surfaces the model already trusts: Wikipedia, Crunchbase, GitHub (for technical brands), industry publications, podcast appearances, third-party reviews, and case-study placements. The more times your brand appears on trusted external surfaces, the more confidently the model resolves your name when a user mentions you.

Pure on-site work has a ceiling. Once your own site is fully optimized, the next gain comes from off-site entity coverage. This is why agencies running serious AEO retainers (including our own AI search optimization retainer and the Calgary-focused Calgary AEO landing) include entity coverage as a discrete workstream.

7. Concrete Claims, Not Marketing Prose

The single biggest content failure we see in audits is marketing prose. Pages full of “we are passionate about delivering excellence” content do not get cited because there is nothing for the model to lift. Citation-grade content is concrete: numbers, timelines, comparisons, definitions, named processes, named people, named outcomes. The test is simple: can a model lift a single sentence from this page that would still make sense as a citation? If not, the page is invisible.

If your site fails any of the seven above, citation share will be limited. If it fails several, citation will be effectively zero. The patterns we see most often:

• OpenAI retrieval agents blocked in robots.txt. Check OAI-SearchBot and ChatGPT-User. Treat GPTBot separately as a training control.
• No JSON-LD schema. The model sees a page but cannot resolve the entity behind it.
• Heavy marketing prose, no concrete claims. Nothing to lift, so nothing to cite.
• No off-site entity coverage. The brand only exists on its own domain, which makes it look thin to the model.
• Slow or JS-heavy pages where content only renders client-side. Do not assume an AI crawler runs JavaScript. If your content only arrives via React after page load, some crawlers may never see it.

Most sites we audit fail at least three of these. The remediation order is consistent: fix robots first, schema second, content third, entity coverage fourth. The free Meridian15 AEO audit is the fastest way to see exactly which patterns are biting your site.

Tracking citation share is the missing discipline for most marketing teams. The classic SEO toolkit (Ahrefs, Semrush, Search Console) does not report ChatGPT citations, because OpenAI does not publish that data. Clicks are visible: ChatGPT adds utm_source=chatgpt.com to referral links, so its traffic shows up in analytics (OpenAI publisher FAQ). The practical options:

Manual Spot Checks

The cheapest method: run ten representative buyer queries through ChatGPT in browse mode and record whether your brand is cited. Repeat monthly. This is crude but it works. Track the queries in a spreadsheet, log the result, and watch the trendline. Anything more sophisticated is icing.

Dedicated AI Citation Tools

A handful of tools have emerged specifically for AI citation tracking: Profound, Otterly.AI, Peec AI, AthenaHQ, Brand24, and Goodie. We compared the six best Perplexity rank tracker tools side-by-side here, and all of them cover ChatGPT citations too. We also run agency-built dashboards as part of the SEO and AEO retainer. The shared shape is the same: scheduled queries across ChatGPT, Perplexity, Claude, and Google AI Overviews, with citation share logged over time. For B2B and DTC brands at scale, this is the real lever.

Our Own AEO audit

The Meridian15 AEO audit is not a citation tracker, but it tells you whether your site is even readable to ChatGPT in the first place. If you are failing the foundational checks, no citation tracker will read positive numbers because the underlying signal is broken. Audit first, then track. The audit is free and runs in seconds.

Getting cited by ChatGPT is not magic and it is not gated by spend. It is a stack of seven concrete moves, six of which most teams can ship in a quarter, and a seventh (entity coverage) that compounds over a year. The brands that ship the stack early get an outsized share of citation real estate while the bar is still low. The brands that wait will spend the next two years catching up with competitors who already became the named source.

If you only do three things this quarter: make the content crawlable, allow OAI-SearchBot and ChatGPT-User where your policy permits retrieval, and publish accurate structured data with accountable authorship. An optional llms.txt and matching FAQPage markup can support discovery and clarity, but neither guarantees citations. If you want to go further, the full retainer playbook lives at SEO and AEO services and the Calgary-specific applied version is at Calgary AEO services. Local Calgary buyers can also read AEO for Calgary businesses for the local-market framing.