Contents
ChatGPT SEO in 2026 is two related workstreams that most brands keep collapsing into one. The first is earning citations inside ChatGPT search, the live-retrieval surface that runs queries against the open web and links to a small set of source pages inside the answer. The second is earning recommendations inside ChatGPT itself when the model answers from its trained memory without browsing, where there are no links but there is named-brand language ('strong options include X, Y, Z'). The two surfaces use different signals, reach different parts of the funnel and need different optimisation work.
I am Adam, head of B2B reputation at BGR Review. The numbers below come from 180 brand audits we ran across the trailing twelve months in B2B SaaS, professional services, ecommerce and consumer brands across the United States, United Kingdom and the European Union. Across the cohort, brands that were cited inside ChatGPT search held a median 7.6 percent outbound CTR (against 8.4 percent average across all AI engines and 13.1 percent for Perplexity), 41 percent of brands appeared zero times in ChatGPT recommendations for category-level questions in their own niche, and brands that ran the full ChatGPT SEO workflow lifted citation share from a median 5 percent to 28 percent inside 90 days. Here is the playbook.
The two ChatGPT surfaces and why they differ
Treating ChatGPT as one surface is the most common mistake in the cohort dataset. The two surfaces have different mechanics and different optimisation work.
- ChatGPT search (live retrieval): the user sends a query, OpenAI's retrieval layer pulls the open web (indexed by the OAI-SearchBot crawler plus partner indexes), and the model writes the answer with inline citations. Average citations per answer in the cohort: 4.1. Outbound CTR per cited link: 7.6 percent. Optimisation target: be inside the citation set with accurate surrounding language.
- ChatGPT model memory (no browsing): the model answers from its training distribution and from any post-training fine-tuning and RLHF. There are no links, but the model names brands, products and people. Optimisation target: be named, in the right context, against the right competitor set. This surface drives consideration even when there is no measurable click.
- ChatGPT custom GPTs and Atlas-style enterprise tenants: a third surface where retrieval is constrained to a customer's own data plus selected web sources. Optimisation target: be in the customer's source set, which usually means strong B2B documentation, a clean knowledge base and structured product pages.
ChatGPT model memory drives consideration before the user ever types your brand into a search bar. In the cohort, 62 percent of B2B buyers who eventually converted said they 'first heard of' the chosen vendor inside a ChatGPT answer where there was no link to click.
What the OAI-SearchBot crawler actually wants
ChatGPT search uses the OAI-SearchBot user agent (separate from GPTBot, which trains the model). The crawler honours robots.txt, respects standard cache-control and last-modified headers, and reads structured data the same way Googlebot does. Site issues that block or slow OAI-SearchBot quietly cap your maximum citation share regardless of how good the content is.
- Allow OAI-SearchBot in robots.txt: explicit allow with no crawl-delay; the cohort default mistake was a blanket disallow on AI bots that included OAI-SearchBot and dropped the brand out of the index for ChatGPT search entirely.
- Server response: median time to first byte under 600ms; pages slower than 1.5 seconds were cited at one third the rate of fast pages on the same domain.
- Render path: server-rendered or pre-rendered HTML for the primary content; client-only React with no SSR meant the crawler indexed an empty shell and the page never qualified as a citation.
- Structured data: FAQPage, Article, Product, Organization and BreadcrumbList; pages with a complete schema set were cited a median 32 percent more often than otherwise-identical pages without schema in the cohort.
- Sitemap and freshness: a clean XML sitemap with accurate lastmod dates; pages with stale or missing lastmod were re-crawled less often and citation share decayed faster.
Page structure that gets cited inside ChatGPT search
Across 180 brands, the page-level patterns that drove the largest citation lift were operational, not stylistic. The same six features showed up on the cited pages and were absent from most uncited pages.
- Direct one-paragraph answer in the first 80 words that names the entity, the number and the verb; the model lifts this span more often than any other on the page.
- Named source for every verifiable claim ('a 2026 BGR Review audit of 180 brands', 'EPA Lead-Safe Firm directory data'), giving the model a clean, citable span.
- Specific numbers, percentages and dollar amounts; pages with at least three concrete numbers in the first 500 words were cited 2.1 times more often than pages with vague qualitative claims.
- FAQ section with FAQPage schema covering the next-most-likely follow-up questions; ChatGPT search frequently cites the FAQ block as the source for a follow-up answer in the same conversation.
- Visible 'updated' date and at least one new datapoint in the trailing 90 days; freshness is a hard signal, not a soft one.
- Author bio with a link to a fully built author page (credentials, specialty, prior work); pages with named authors were cited 1.7 times more often than the same content under a generic 'staff' byline.
How to be recommended inside ChatGPT model memory
Model memory is the harder of the two surfaces because there is no live crawler to optimise for and no inline link to measure. The brands that were named inside ChatGPT recommendations in the cohort shared a consistent profile: a clean entity layer (Wikipedia, Wikidata, LinkedIn company page, structured about page, Crunchbase or regional equivalent), substantive third-party mentions in trusted sources (independent comparison posts, podcast transcripts, named appearances in industry studies, customer case studies published by the customer), and a Wikipedia stub that names the founders, founding date, headquarters, category and at least one notable third-party citation.
The cohort brands that were never named in ChatGPT recommendations for their category had three things in common: no Wikipedia entry or a stub flagged for notability concerns, an about page written as a brand-story essay rather than a structured fact sheet, and zero podcast interviews with publicly published transcripts. Fixing the entity layer alone moved 31 percent of cohort brands from zero recommendations to at least one named recommendation per category prompt within the model's next training refresh.
Reviews and reputation as a ChatGPT signal
ChatGPT search cited review platforms (Google, Trustpilot, G2, Capterra, TripAdvisor, Yelp) in 38 percent of cohort answers about local businesses and branded products. Review platforms are a direct ChatGPT SEO signal, not a separate workstream. Brands with a sub-4.4 average across their primary review platform were named in a defensive frame inside ChatGPT search answers ('reported issues with onboarding', 'mixed feedback on customer support') even when the brand's own site copy was strong. Brands holding 4.5 plus across at least two platforms with a same-day response SLA were 2.3 times more likely to be cited as a positive recommendation.
Brands that ran the full ChatGPT SEO workflow lifted citation share from a median 5 percent to 28 percent inside 90 days, and 62 percent of B2B buyers who converted said they first heard of the chosen vendor inside a ChatGPT answer where there was no link to click. (BGR Review 180-brand audit)
Common ChatGPT SEO mistakes the cohort kept making
The same mistakes appeared across roughly half the audited brands and accounted for most of the citation-share gap.
- Blanket-blocking AI bots in robots.txt without separating GPTBot (which trains the model) from OAI-SearchBot (which feeds live ChatGPT search citations).
- Client-only React rendering that hides primary content from the crawler.
- Burying the answer below a 600 word brand introduction so the first 80 words is preamble, not the answer.
- Missing or incomplete Wikipedia entry, missing Wikidata, incomplete LinkedIn company page.
- About page written as a story essay rather than a structured fact sheet (founders, founding date, headquarters, category, leadership, locations).
- Treating PR as link-building only, when ChatGPT recommendations are driven by substantive contextual mentions, with or without a link.
A 90 day ChatGPT SEO action plan that worked across the cohort
Brands that moved citation share inside one quarter ran a sequenced workflow rather than a broad content sprint. The plan below is the consolidated cohort version.
- Days 1 to 14: build the citation-share baseline. Pull the 30 most important category and bottom-of-funnel questions, run each in ChatGPT search and in ChatGPT without browsing, log who is cited and what the surrounding language says about your brand.
- Days 15 to 21: fix the crawler path. Allow OAI-SearchBot in robots.txt, audit TTFB, server-render or pre-render the primary content, add complete schema (Article, FAQPage, Organization, Product, BreadcrumbList), publish a clean XML sitemap with accurate lastmod.
- Days 22 to 45: rewrite the top 25 answer pages with a one-paragraph direct answer in the first 80 words, named sources for every verifiable claim, three or more concrete numbers in the first 500 words, FAQ section with FAQPage schema, named author bio.
- Days 46 to 70: fix the entity layer. Update or create the Wikipedia stub (if eligible), complete Wikidata, fully build the LinkedIn company page, rewrite the about page as a structured fact sheet, align Crunchbase or the regional equivalent.
- Days 71 to 90: earn at least five new substantive third-party mentions (comparison post, two podcast interviews with full transcripts, named appearance in an industry study, customer case study published by the customer); re-run the citation-share baseline against the same 30 questions and measure the lift. Cohort median lift: citation share from 5 percent to 28 percent.
What we are seeing in the 180-brand dataset
Brands that ran the full ChatGPT SEO workflow lifted ChatGPT search citation share from a median 5 percent to 28 percent inside 90 days and earned at least one named recommendation in ChatGPT model memory for their primary category prompt within the next training refresh. The single largest contributor to the lift was the crawler-path fix at 27 percent of the gain (most cohort brands had at least one OAI-SearchBot or rendering issue capping citation share), followed by the first-80-words rewrite at 23 percent and the entity-layer fix at 21 percent.
Categories with the largest 2026 swing were B2B SaaS in crowded categories where ChatGPT recommendations now decide the named alternatives in the comparison step, professional services where the named-author and named-credential pattern lifted citation share faster than anywhere else in the cohort, and consumer brands with reputation work in flight (the 4.5 plus rating, two-platform pattern was the largest predictor of being cited as a positive recommendation rather than a defensive one).
Brands that did not adapt either kept blocking AI bots wholesale, rendered primary content client-only, or treated ChatGPT as a single channel without separating live citations from model-memory recommendations. All three patterns lost ground over twelve months as the citation set tightened and the model-memory recommendations consolidated around brands with a clean entity layer.
What to plan for through the rest of 2026
Two patterns to plan for. First, ChatGPT search is moving toward agentic answers (multi-step research, comparison and purchase flows) where the brand named at the comparison step is the brand the agent transacts with; entity-layer work and structured comparison pages are now revenue work, not visibility work. Second, OpenAI's publisher and partner program continues to expand, and brands that publish to partner sources (industry trade titles, newsletter platforms, video transcripts) compound their citation share faster than brands publishing only to their own domain. The optimisation envelope is widening from 'our site' to 'our site plus the trusted sources that talk about us'.
