Yes. You can get recommended by ChatGPT, and businesses do it every day. By late 2026 there’s enough published evidence to say how. What you can’t do is buy it, hack it, or shortcut it with a magic file, because ChatGPT recommends brands it can verify across the sources it actually reads.
Work on getting mentioned in the sources ChatGPT already cites for your category. Check that your product details agree across your site, review profiles, and comparison pages. Give those sources specific facts they can verify.
Everything else (llms.txt worship, schema obsession, “guaranteed ChatGPT placement” pitches) has little or no evidence behind it. With ChatGPT now reporting over a billion weekly users, it’s worth getting this right. So let’s separate what works from what wastes your time.
One definition first, because the jargon gets in the way: generative engine optimization is the practice of making your business easy for AI models to find, verify, and recommend, as opposed to classic SEO, which tries to rank pages. In practice, most of it happens off your website, in places you don’t control.
How ChatGPT actually decides what to recommend
ChatGPT blends three inputs when it recommends a business, and none of them is your keyword density.
Training data. OpenAI’s GPTBot crawls the web to train its models. When thousands of independent sources describe you the same way, that consensus becomes what the model “knows” about you.
Live search. Since the ChatGPT search launch in October 2024, ChatGPT has pulled live web results with a fine-tuned GPT-4o model that relies on “third-party search providers, as well as content provided directly by our partners.” Answers ship with a Sources panel so users can check where claims came from. By February 2025 it was available to everyone, no signup required.
Partners and feeds. OpenAI licenses content from publishers like Reuters, News Corp, and The Atlantic, and it built dedicated surfaces for shopping. Shopping research, launched in November 2025, asks clarifying questions and builds buyer’s guides from retail sites, and OpenAI states plainly that “results are organic and based on publicly available retail sites.” In March 2026, OpenAI extended its Agentic Commerce Protocol so merchants can share product feeds and promotions, “so their catalogs are fully represented in ChatGPT.” If you sell products, that feed is now a real distribution channel.
There’s also a blunt technical gate: OpenAI’s crawler documentation says that if you block OAI-SearchBot in robots.txt, your site “will not be shown in ChatGPT search answers.” Changes take about 24 hours to take effect. That’s the cheapest fix in this article, and I still find sites that fail it.
Put those three inputs together and the implication is uncomfortable but clarifying: ChatGPT’s opinion of your business is mostly an average of everyone else’s. Your website is one voice in a choir. If the choir says you’re a solid mid-market option for a specific buyer, that’s what comes out, and no amount of on-page polish overrides a chorus of contradicting sources.
What the evidence actually says
Most AI search optimization advice is old SEO advice wearing a new jacket. Rather than add to the pile, I put the main tactics next to the published evidence, and left out anything I couldn’t source.
| Tactic | Evidence status | What the evidence says |
|---|---|---|
| Broad brand mentions across the web | Supported | Ahrefs’ 75,000-brand study (May 2025) found branded web mentions had the strongest correlation with AI Overview visibility (0.664); backlinks managed 0.218. 26% of brands had zero AI mentions. |
| Presence on review and directory sites | Supported | BrightLocal (Aug 2026) analyzed roughly 2 million citations across ChatGPT and Google AI tools: Yelp appeared as a source in 80% of ChatGPT’s local answers. |
| Citations, quotes, and statistics in content | Supported, with caveats | The Princeton GEO paper (KDD 2024) measured visibility gains up to 40% from its optimization methods, varying by domain and starting position. |
| Allowing AI crawlers | Supported (necessary, not sufficient) | OpenAI’s docs: blocking OAI-SearchBot means your site won’t appear in ChatGPT search answers. |
| Product feeds via ACP | Supported for shopping surfaces | OpenAI (Mar 2026): merchant feeds get catalogs “fully represented”; Shopify Catalog data flows in automatically. |
| Schema markup as an AI-citation lever | No reliable evidence | Ahrefs’ matched experiment (May 2026, 1,885 pages) found ChatGPT citations moved +2.2% after adding JSON-LD, statistically indistinguishable from zero. |
| llms.txt | No reliable evidence | Ahrefs (June 2026): 97% of roughly 38,000 llms.txt files were never fetched, not once, in a full month; just 1.1% of all requests came from AI retrieval bots. |
| Paying to be recommended | Not possible as advertised | OpenAI began testing ads in ChatGPT with US Free and Go-tier users in January 2026, but ads are labeled placements beside answers, not the model’s recommendation. |
Ahrefs describes these results as correlations. But when independent teams with different methods keep landing in the same direction, I’m comfortable acting on it. You should be too.
You can also test on your own, which I’d recommend before spending a dollar. Write ten questions your ideal customer would actually ask, run them weekly in a fresh chat, and log which brands show up. Track your mention rate alongside your presence on the sources cited. It’s crude, but it beats trusting a vendor dashboard that promises certainty this space can’t deliver yet.
What works, in the order I’d do it
-
Fix crawler access first. Check robots.txt for blocks on GPTBot, OAI-SearchBot, and ChatGPT-User, then confirm your key pages are indexable in the first place. If you blocked AI bots during the 2023 panic, unblock them, the change takes effect in about a day. Blocking doesn’t guarantee invisibility everywhere, but it does guarantee exclusion from ChatGPT’s search answers.
-
Make your entity consistent everywhere. One name, one category description, one list of services, repeated across your site, profiles, directories, and social accounts. Models resolve businesses by consensus, so conflicting facts lower their confidence. Seer Interactive’s analysis of a 10,000-question LLM dataset (April 2025) found branded search volume was among the strongest correlates of AI mentions (second only to domain rank) which tracks with what I see testing my own products.
-
Show up on the sources your category’s answers actually pull from. BrightLocal’s August 2026 study ran 60,970 checks across the US, UK, and Australia and found Yelp cited in 80% of ChatGPT’s local answers, followed by Tripadvisor, OpenTable, BBB, and Angi. Google’s AI tools leaned hardest on Google Business Profile at 28.5% of all citations. Notably, ChatGPT drew on only about 6,000 unique domains across all those checks, the list you need to be on is short, so find the short list for your vertical and get on it.
-
Take reviews seriously, volume, recency, and specificity. Review platforms dominate citations in local AI answers, and models often quote review content when they explain a recommendation. A steady flow of recent, detailed reviews beats a wall of five-star one-liners from 2019. Reviews and AI recommendations are linked now the way reviews and local pack rankings were linked a decade ago, ignore that at your peril.
-
Publish specific, verifiable facts that answer buying questions. The Princeton GEO paper (KDD 2024) tested nine optimization methods and found that adding citations, quotations, and statistics ranked among the most effective. Translation: real prices, real specs, real comparisons, real tradeoffs. “We’re the best” is noise to a model; “here’s what we cost, who we fit, and where we fall short” is signal it can quote. If a page wouldn’t survive being pasted into a chat as a standalone answer, rewrite it until it would.
None of this is exotic. It’s mostly unglamorous entity hygiene plus content people actually need, applied relentlessly.
What doesn’t work (or has no evidence yet)
- llms.txt as a visibility switch. Ahrefs’ June 2026 study of 137,210 domains found 28% publish an llms.txt file, and 97% of those files were never fetched, not once, in May 2026. Of the requests that did happen, only 1.1% came from AI retrieval bots. Ahrefs’ verdict: “largely decoration.” Add it if you like; just don’t expect anything from it.
- Schema as your GEO tactic. Ahrefs tracked 1,885 pages that added JSON-LD between August 2025 and March 2026, each matched against control pages. ChatGPT citations moved 2.2%, indistinguishable from zero. Schema still earns rich results in classic search; it’s just not an AI visibility lever.
- Paying to be recommended. You can buy labeled ads in ChatGPT, but there is no paid inclusion in the model’s actual answers. Anyone selling “guaranteed ChatGPT recommendations” is selling ads or vapor. Ask them which.
- Prompt-injection tricks. Researchers including a Harvard team (April 2024) showed adversarially optimized text can boost a product’s visibility in LLM recommendations. That’s an attack, not a marketing strategy. It’s the kind of thing platforms patch out and punish, and it treats your customers as collateral.
- Blocking AI bots while expecting AI visibility. See OpenAI’s docs above. If OAI-SearchBot can’t read you, ChatGPT search can’t show you.
- Submitting to “AI directories.” There is no registry that gets you recommended, and no file you upload that flips a switch. Google’s own AI features documentation says you don’t need “new machine readable files, AI text files, or markup” to appear in its AI features, and ChatGPT works the same way, reading the same web everyone else does.
What’s still unknown
Answers are volatile. The same question can produce different businesses on different days, and model updates reshuffle recommendations without warning. Any single audit snapshot is a photo, not a movie.
Most evidence is correlational. Ahrefs’ 0.664 brand-mention correlation doesn’t prove mentions cause visibility, big brands do everything more. The Princeton numbers come from a benchmark, not your market.
The retrieval corpus is opaque. Nobody outside OpenAI knows exactly which pages and feeds a given answer can draw from, and personalization through memory means two users can get different answers to the same question.
Durability is unproven. My bet (and it’s a bet) is that consensus-driven visibility persists across model updates better than tricks do, precisely because consensus is what models are built to converge on. Nobody has published proof of that yet. Treat “future-proof” claims accordingly, including mine.
FAQ
Can you influence ChatGPT recommendations? Yes. Not with prompt hacks or hidden files, but by being easy to verify: consistent entity information, presence on the review sites and directories models read, and content with specific, checkable facts. The evidence for those inputs is solid; the evidence for tricks isn’t.
What makes ChatGPT recommend a brand? Consensus across the sources it reads. In local answers that’s review platforms and business data; in shopping answers it’s retail pages and product feeds; everywhere it’s broad mentions across the web. Ahrefs found web-wide brand mentions correlate far more strongly with AI visibility (0.664) than backlinks do (0.218).
Do reviews affect AI recommendations? Yes, especially locally. BrightLocal’s August 2026 study found Yelp appeared as a source in 80% of ChatGPT’s local answers, with Tripadvisor, OpenTable, BBB, and Angi also cited. Volume and recency matter, but so does detail, models quote specific review content when justifying picks.
Does llms.txt work? No evidence says it does. Ahrefs found 97% of llms.txt files were never fetched in a month, and AI retrieval bots accounted for about 1% of the requests that did happen. Google likewise says no special AI files are needed for its AI features. It’s harmless, not helpful.
Can you pay ChatGPT to recommend you? No. OpenAI started testing ads in ChatGPT in January 2026 for Free and Go-tier users, but those are labeled sponsored placements. They don’t change the model’s organic recommendations, and paid subscribers don’t see ads at all. There’s no paid inclusion in the answer itself.
How long does it take to show up in AI answers? Two timelines. Live search results can surface you within weeks of your site being crawlable and cited elsewhere. Training-data recognition takes months, because it only updates when OpenAI trains new models. The 24-hour robots.txt change is the only fast lever OpenAI documents, everything else compounds slowly.
Sources
- OpenAI (Introducing ChatGPT search) https://openai.com/index/introducing-chatgpt-search
- OpenAI (Introducing shopping research in ChatGPT) https://openai.com/index/chatgpt-shopping-research
- OpenAI (Powering product discovery in ChatGPT) https://openai.com/index/powering-product-discovery-in-chatgpt
- OpenAI (OpenAI crawlers and bots) https://developers.openai.com/api/docs/bots
- Google Search Central (AI features and your website) https://developers.google.com/search/docs/appearance/ai-features
- Ahrefs (An Analysis of AI Overview Brand Visibility Factors (75K Brands)) https://ahrefs.com/blog/ai-overview-brand-correlation
- Ahrefs (Does Adding Schema Get Your Pages Cited by AI?) https://ahrefs.com/blog/schema-ai-citations
- Ahrefs (We Analyzed 137K Sites: 97% of llms.txt Files Never Get Fetched) https://ahrefs.com/blog/llmstxt-study
- BrightLocal (Where to get local citations for AI search) https://www.brightlocal.com/resources/ai-directory-sources
- Seer Interactive (Does Brand Awareness Impact LLM Visibility?) https://www.seerinteractive.com/insights/does-brand-awareness-impact-llm-visibility
- Aggarwal et al. (GEO: Generative Engine Optimization (KDD 2024)) https://arxiv.org/abs/2311.09735
- Kumar & Lakkaraju (Manipulating Large Language Models to Increase Product Visibility) https://arxiv.org/abs/2404.07981
