How to get your B2B SaaS cited by ChatGPT: a 90-day plan

Getting cited by ChatGPT is a sequencing problem, not a tactics problem. You have to be describable first, then retrievable, then corroborated, and only then can you measure movement. Each phase takes weeks, and the full loop runs about 90 days. Most of that time is the re-probe interval, not the work.

If ChatGPT just named a competitor for the problem you solve, here is what happened: the model could resolve who they are in one sentence, and it could not do the same for you. That is the gap. This plan closes it in order. (For the mechanics underneath it, start with what is GEO.)

Why does getting cited take 90 days and not a weekend?

Because the bottleneck is not your content. Only about 18% of ChatGPT conversations trigger a live web search (a Search Engine Land study, as of June 2026); a brand that exists only on its own site is invisible to roughly four answers in five. The rest answer from model weights, which update on a training cycle, not in real time. A page you publish today may not influence model weights for weeks or months. What you can influence faster is the off-site footprint that feeds both the crawlers and the third-party sources models already trust.

The second reason is measurement. A single re-probe cannot tell signal from noise. You need a frozen prompt panel, at least three runs per prompt per engine, and a reported interval, not a point estimate. That takes three check-ins at days 30, 60, and 90.

The third reason is sequencing. Skipping phase 1 (identity) and going straight to content or outreach is the most common mistake. A score of zero on an AI visibility audit usually means the engine cannot resolve who you are, not that it judged you and moved on. You cannot be recommended for a category if the model cannot first answer “what is this company.”

Days 1–14: Can an AI say what you are in one sentence?

Write this sentence right now: “[Product] is a [category noun] for [who], built by [company].” That is your canonical description. It should appear verbatim in three places before day 14: above the fold on your homepage, on your about page, and in the bio field of your LinkedIn company page.

That is the whole of day one, and it takes under an hour.

Beyond the sentence, add Organization schema markup to your homepage: a structured-data block that gives crawlers your name, URL, and description in a format they parse without reading prose (the how is covered in schema markup for AI visibility). Then confirm the same name appears on your domain, your LinkedIn page, your Crunchbase profile, and any press mentions. Inconsistent naming is the most common cause of a zero-resolution result.

The goal of this phase is narrow: an AI assistant asked “what is [your product]” should return a one-sentence answer. Nothing else in the plan works until that is true.

Days 15–45: Does a page exist that answers each of your five buyer questions?

Once you are describable, you need to be retrievable. That means a dedicated page for each of the five questions a buyer asks before they sign up: what is it, who is it for, how does it work, what does it cost, and how does it compare to the alternative they already know.

Each page needs three things to be liftable by a model:

• An answer-first paragraph. The direct answer appears in the first two sentences, not after context-setting.
• At least one concrete number, named example, or dated claim per section. Princeton’s KDD 2024 study across 10,000 queries found that statistics, quotations, and cited sources each lift AI citation rates by 30 to 41 percent. Vague claims lift nothing.
• A structure the model can extract from. A numbered list or a definition block outperforms a wall of prose on evaluative queries.

One page per question. Do not combine them. A page titled “Features and Pricing and FAQ” answers none of those questions well enough to be cited for any of them.

If you are a team of one, write the “who is it for” and “how does it compare” pages first. Those are the pages buyers read at the evaluation stage, and the pages models cite when someone asks “what tool should I use for [problem].” For what makes a page citable, see three signals that lift AI citation rates.

Days 30–75: Who else says it?

Models are roughly 6.5 times more likely to cite a third-party source than a brand’s own domain on evaluative queries, and about 85% of brand mentions in AI answers come from third-party pages (as of June 2026, from converging vendor crawl data). Your own pages are necessary but not sufficient.

The third-party sources that move evaluative queries, in rough priority order:

• Software review directories where buyers in your category already search. Claim your listing, fill every field, and ask 5 to 10 customers for a review this month.
• Comparison and alternative pages on sites that already rank for your category. A mention on a “best tools for [problem]” list on a domain the model already trusts is worth more than a new page on your own site.
• Community threads on Reddit, Hacker News, or a Slack community where your buyers congregate. A genuine answer to a question, with your product named and described accurately, is a citable source.
• Partner and integration pages. If you integrate with a tool your buyers already use, ask that team to list you on their integrations page with your canonical description.

What to ask for: a named mention with your canonical one-sentence description, a link to your homepage or the relevant product page, and ideally a use-case sentence. A generic “check out [product]” mention does not give a model enough to work with.

This phase overlaps with phase 2 because some of the pages you publish in days 15–45 will attract third-party mentions on their own. Start the outreach at day 30 regardless.

Days 30, 60, and 90: How do you know it moved?

Freeze your prompt panel on day zero: write down the exact prompts you will use to check visibility, the exact engines, and the date. Do not change the prompts between check-ins. A prompt that shifts between runs is not a measurement, it is a new experiment.

At each check-in, run each prompt at least three times and report the range, not a single result. If your product appears in 1 of 3 runs, that is a wide-interval maybe. If it appears in 3 of 3, that is closer to a reliable signal, though still a small sample.

Track per prompt, per engine: whether you are mentioned, whether the mention is accurate, and whether a competitor is mentioned instead. That last column tells you whether you are losing a specific framing, not just absent overall.

One anonymized receipt from this exact sequence: after shipping the identity and corroboration phases across its GitHub and docs surfaces, a Series B devtools company’s citation rate moved from 1.25% to 5.83% in six weeks on its frozen panel. One brand, one panel; a receipt, not a promise.

For the measurement setup in full, see how to measure AI visibility properly.

What usually moves first, and what does not move in 90 days?

What moves first: definitional queries. “What is [product]” and “what does [product] do” respond fastest to phase 1 work because they require only identity resolution, not corroboration.

What moves in the middle: category queries. “Best tools for [problem]” and “alternatives to [competitor]” start moving after phase 3 outreach lands on two or three trusted third-party pages.

What does not move in 90 days: high-competition evaluative queries where established players have years of third-party corroboration. If a competitor has 400 reviews on a major directory and you have 12, that gap does not close in a quarter. The 90-day plan gets you into the conversation; it does not guarantee a top position on the hardest queries, and anyone who promises one is selling you a position they do not control.

How many hours a week does this take for a team of one?

• Phase 1 (days 1–14): roughly 4 to 6 hours total. One hour for the canonical sentence and profile updates, 2 to 3 hours for schema markup, 1 to 2 hours to audit name consistency.
• Phase 2 (days 15–45): roughly 3 to 5 hours per page across 5 pages, so 4 to 6 hours per week for four weeks.
• Phase 3 (days 30–75): roughly 2 to 3 hours per week for outreach, follow-up, and tracking placements.
• Check-ins (days 30/60/90): 1 to 2 hours each.

Call it 50 to 70 hours across the quarter. Weeks 6 through 12 are mostly waiting and re-probing, not building. That is by design: the re-probe interval is part of the work. If you want the same loop run hands-on with your team instead, that is what the 90-day sprint is.

Day zero is the free scan. It tells you which of the four phases you are actually stuck in, so you spend the next 90 days on the right problem, not the visible one. Run the free scan; it takes about 90 seconds.

---

Provider behavior changes quickly. This guide reflects what we know as of August 2026; we update it when the evidence shifts. Numbers carry their sources because we believe in showing the work.