How to get your brand mentioned by ChatGPT

ChatGPT reached roughly 900 million weekly active users in early 2026, about double the figure from a year earlier. That scale alone would make it worth watching, but the more important shift is who those users are and what they are doing. G2's 2026 buyer research found that 51 percent of B2B software buyers now start their research with an AI chatbot more often than with Google, up from 36 percent just seven months earlier. Within that group, 72 percent use ChatGPT specifically during vendor evaluation, making it the dominant chatbot for software research at 63 percent share.

Forrester's 2026 Buyers' Journey Survey of 18,000 business buyers puts an even larger number on the trend: 94 percent used AI at some point during their most recent purchase, with 55 percent comparing vendors inside AI tools and 47 percent building internal business cases before ever contacting a vendor. The shortlist is increasingly assembled before your sales team knows the deal exists.

The consequences are measurable. In the same G2 research, 69 percent of buyers chose a different vendor than they originally expected because of what an AI chatbot told them, and 33 percent bought from a brand they had never heard of before the AI named it. Meanwhile, 51 percent of B2B tech brands have zero citations across ChatGPT, Perplexity and Gemini. Half the market is invisible in the channel where a third of buyers discover their eventual vendor. That gap is the opportunity this playbook is built to close.

Before touching tactics, understand the two separate systems that produce a ChatGPT answer, because each responds to different work. The first is training data: the model's baked-in knowledge of brands, products and reputations, frozen at the point the model was trained. If your brand appeared widely and positively in that corpus, ChatGPT can recommend you with search turned off. You cannot schedule when that knowledge updates, because OpenAI does not publish retraining timelines.

The second system is live retrieval. When a question benefits from current information, ChatGPT searches the web, reads a handful of pages, and composes an answer with citations. ChatGPT's search experience launched on Bing's index, but OpenAI now operates its own crawler, OAI-SearchBot, and has roughly tripled its web crawl since August 2025, according to a Botify analysis. This retrieval layer is the part you can influence in weeks rather than years, and it is where most of this playbook operates. The distinction matters enough that we wrote a full explainer on training data versus live retrieval.

A practical test: ask your target question in ChatGPT and check whether the answer carries citations. Cited answers came through search and can change as soon as the retrieved pages change. Uncited answers lean on training data and will move slowly no matter what you publish this month. For a deeper look at the machinery, see how ChatGPT decides which brands to recommend.

Start with a list of 20 to 25 questions your buyers actually type when they shop your category. Good sources: your sales team's discovery calls, support tickets that begin with a comparison, the questions section of competitor reviews, and the phrasing your customers used before they found you. Write them the way a buyer would, in plain language, and keep them unbranded. "Best invoicing tool for freelancers" tells you whether ChatGPT recommends you to strangers; "Is Acme good for invoicing" only tells you it can echo your own name back.

Then run each question in ChatGPT, and run it more than once. This is the step most manual audits get wrong. SparkToro measured under a 1 percent chance that two identical ChatGPT runs return the same brand list, so a single pass gives you an anecdote rather than a baseline. Three to five runs per question, spread over a few days, is the minimum for a signal you can trust. Log every brand named, whether yours appears, and which sources the answer cites.

Done by hand in a spreadsheet, this takes the better part of a day for 25 questions and needs repeating every few weeks. This is the part Reachroller was built to automate: it runs your question set through the official ChatGPT API on a schedule, stores every raw answer so you can audit the score, and excludes branded questions from the headline number so the baseline is honest. One credit equals one AI answer, and the methodology is documented in full. Either way, do not proceed to writing until you know precisely which questions you lose and to whom.

For every question you lose, open the citations in the answers you logged. The patterns will surprise you if you carry SEO instincts in. According to 5W Research's 2026 analysis, Wikipedia accounts for 13.15 percent of ChatGPT citations in the U.S. and Reddit for 11.97 percent, which means two sites supply over a quarter of everything ChatGPT cites. The Wall Street Journal, the New York Times and Bloomberg do not appear in the top 20. Prestige with humans and weight with the engine are different things.

Beyond the two giants, the citation graph fragments into a long tail of niche roundups, comparison posts, community threads and reference pages, a pattern confirmed by both Semrush's three-month most-cited-domains study and Otterly's 2026 AI Citations Report, which drew on more than a million data points. Your job in this step is to build a target list: the specific pages and domains ChatGPT cites for your questions today. Those are the doors worth knocking on in step 5, and the formats worth imitating in step 3. The full breakdown lives in what ChatGPT actually cites.

One caution while you build the list: what works for ChatGPT does not automatically transfer. Cross-platform citation analyses find only about 11 percent of domains are cited by both ChatGPT and Perplexity. Treat this step as engine-specific research, and expect to repeat a lighter version of it for each engine you decide matters to your brand.

Now write the page that answers a lost question better than whatever ChatGPT currently cites. This is the one part of AI visibility with peer-reviewed evidence behind it. The Princeton and Georgia Tech GEO study, published at KDD 2024, tested nine optimization methods across a large query benchmark and found that adding quotations, statistics and cited sources boosted a page's visibility in generative engine responses by up to roughly 40 percent. The best methods improved about 22 percent on position-adjusted word count and about 37 percent on subjective impression versus baseline. The same study found keyword stuffing performed near the bottom, worse for generative engines than doing nothing.

Structure the page for extraction. Use the buyer's question, or a close paraphrase, as the h1. Open with a 90 to 130 word paragraph that answers the question completely and can stand alone when quoted, because that opening paragraph is exactly what an engine lifts. Follow with sections that carry real numbers attributed to named sources, direct quotations where you have them, and an honest FAQ. Every claim should survive a skeptical reader, because the engine is choosing between your page and pages that already earned its trust. Section-by-section guidance is in how to write content AI engines actually cite.

Writing one page this way takes hours; writing one for each of 15 lost questions is a quarter of content work. Reachroller generates each fix as a publish-ready page, with the URL slug, title tag, meta description and schema markup included, at a cost of ten credits per fix. You review, edit and publish under your own domain. The generator applies the GEO findings by default: answer-first opening, attributed statistics, quotable structure. What it will not do is invent facts about your product, because a fabricated claim that gets cited is a liability, and every generated draft deserves a human read before it ships.

A page that is not in a search index does not exist for retrieval-based answers. ChatGPT's search experience launched on Bing's index, which makes Bing Webmaster Tools a first-class citizen in this playbook rather than an afterthought, and OpenAI's own OAI-SearchBot increasingly crawls the web directly. Submit the new page in both Google Search Console and Bing Webmaster Tools, confirm it appears in each index with a site: query, and check your server logs or analytics for OAI-SearchBot visits if you want direct evidence that OpenAI has fetched it.

Also confirm you are not blocking the crawlers you need. Some robots.txt files written during the 2023 wave of AI blocking still deny every AI user agent, including the search crawlers that produce citations and referrals. Blocking training crawlers is a defensible choice; blocking OAI-SearchBot removes you from consideration for cited answers entirely. Audit the file before you wonder why nothing moves.

Then give it time, and be suspicious of anyone who says otherwise. Indexing plus engine pickup runs one to two weeks in the normal case. Recheck earlier and you will conclude the work failed when it has not been tested yet. This waiting period is also why Reachroller schedules its recheck rather than running it immediately after you publish: the loop only closes when the interval has been honest.

Your own page is half the battle. Citation analyses from Ahrefs and Semrush consistently show AI engines leaning on independent third-party pages, reviews, roundups and reference sites, more heavily than on brand-owned domains. Getting your brand named on one page the engine already cites for your question often moves the answer faster than anything you publish yourself. This is where the target list from step 2 pays off: you know exactly which roundup, which comparison post, which community thread the engine reads. Pitch those specifically, with something genuinely useful to add, rather than spraying generic PR.

Reddit deserves its own line. It supplies 11.97 percent of ChatGPT citations per 5W Research, and its citation share in commercial categories grew roughly 73 percent across 2025 and 2026. The temptation to astroturf is real and the penalty is severe: bans, public callouts, and threads about your fake reviews that then get cited by the same engines. The durable play is authentic participation, answering questions in your niche as a named brand representative, which we cover in a Reddit strategy AI engines respect.

Wikipedia, at 13.15 percent of ChatGPT citations, is the other giant, and the most dangerous to approach carelessly. Its notability standards and conflict-of-interest rules are strict, and a clumsy attempt at self-promotion tends to end in deletion and a public log of the attempt. If your brand has genuine third-party coverage, a neutral, well-sourced presence is worth pursuing patiently. If it does not, earn the coverage first. Wikipedia reflects reputation; it does not create it.

After the one to two week interval, re-run every target question the same way you ran the baseline: multiple runs per question, every answer logged. You are looking for a shift in the trend, your brand appearing in three of five runs where it appeared in zero, rather than a binary flip. Because answers are probabilistic, a single triumphant screenshot means as little as a single discouraging one. Compare distributions, then keep the loop running: publish the next fix, pitch the next third-party page, recheck again.

Keep the measurement honest as the numbers improve, because this is where self-deception creeps in. Count a mention only when your brand name literally appears in the answer text. Keep branded questions out of the headline score. Store the raw answers so anyone can audit the number, including you in three months. These principles are the core of Reachroller's scoring, and they are worth adopting even if you never use the product. The full method is in how to measure AI visibility without lying to yourself.

Expect partial wins. Some answers flip in two weeks. Some flip for two runs and revert. Some sit unmoved because the mention lives in training data and no retrieval happens for that question. The loop still compounds: every citable page you publish and every third-party mention you earn raises the probability that the next retrieval lands on you, across every engine that reads the same web.

Keyword stuffing. The GEO study measured it directly: for generative engines it performed near the bottom of all nine tested methods, worse than leaving the page alone. The engine reads for substance it can quote, and repetition of the target phrase gives it nothing. If your SEO muscle memory says to sprinkle the keyword twelve more times, this is the channel where that instinct actively hurts.

Betting on llms.txt. The proposed standard for describing your site to language models is cheap to add and harmless, but no engine has confirmed using it, and no study shows a citation effect. Add it in ten minutes if you like optionality; do not let it displace an hour of work on content or indexing, which have evidence behind them.

Expecting schema to carry the load. Ahrefs' May 2026 study of 1,885 already-cited pages found adding JSON-LD produced no measurable citation lift for ChatGPT. Ship schema because it is low-cost and may help discovery and parsing, but the citation decision is made on the substance of the page. Finally, skip any service promising guaranteed or overnight ChatGPT mentions: the mechanics above are the mechanics, and none of them can be bought outright or completed in a day.