GEO Strategy: A 90-Day Plan for B2B Companies
A GEO strategy is a plan for generative engine optimization: getting your company named and cited when buyers ask ChatGPT, Perplexity, Claude or Gemini for recommendations. A working one has three parts over 90 days: find the questions your buyers actually ask, find the sources those answers pull from, then get placed on those sources and re-test every week.
A GEO strategy is a plan for generative engine optimization. Here is the 90-day version I run: prompt discovery, citation gap analysis, source engineering, and how to measure it.
On this page
Note on verification: every number below was opened at its original source on 17 September 2026, and each one carries a link and a date at the bottom. Two caveats. GeoCited's 40-prompt citation study is my own agency's research, published with its raw data but not independently reviewed, so weigh it as vendor data. And RESONEO's retrieval splits come from one corpus collected in July 2026, with the researchers' own note that the mix shifts by ChatGPT tier, so treat them as a snapshot rather than a constant.
A GEO strategy here means generative engine optimization, not geopolitical strategy.
What makes a generative engine optimization strategy different from an SEO plan
Most of the work is not new. RESONEO analysed 1,249 ChatGPT answers in July 2026 and found that roughly one retrieved URL in three already sits on Google's first page, against one in twenty on Bing's. Classic ranking still feeds the retrieval layer, so a site with no search foundation has nothing for these engines to find.
Two things are actually new. The first is prompt research: your target is a set of buyer questions, not a keyword list. The second is measurement, which behaves nothing like rank tracking.
There is also a retrieval detail worth planning around. In the same RESONEO corpus, pages the model actually opened were cited 74% of the time, while pages that were only retrieved were cited 7% of the time. Being findable and being cited are two different outcomes.
So the plan below is not a content calendar. It is an order of operations.
Days 1 to 14: Sourced Prompt Discovery
The first process is what I call Sourced Prompt Discovery. It builds the question list from your own sales calls and support tickets, then tracks each question across ChatGPT, Perplexity, Gemini and Claude, recording whether you come back named, cited, or absent.
The word "sourced" is the whole point. Questions invented in a keyword tool tend to be the questions nobody asks an AI tool out loud.
One run per question is not a baseline. SparkToro's research, published 28 January 2026 by Rand Fishkin from 600 volunteers and 2,961 runs, found less than a 1 in 100 chance that two answers list the same set of brands. Any single screenshot of an AI answer is close to noise.
That is why the count matters more than the ranking. Ask each question enough times, across enough engines, that you can talk about the share of runs you appear in rather than a position. An AI visibility audit is one way to get that starting count.
Finish the two weeks with a written list: every question, every engine, the share of runs you were named in, and which competitors showed up instead.
Days 15 to 30: Citation Gap Analysis
The second process looks past the brand names to the URLs. Citation Gap Analysis collects every page the engines cited across your question set, then ranks the domains by how often they feed answers where a competitor gets named and you do not.
This is where an abstract generative engine optimization strategy turns into a list of specific pages to build or get onto.
Page type tracks question type more tightly than most people expect. Across 40 buying-intent prompts and 490 cited URLs in my own agency's study on 28 August 2026, comparison pages led "X vs Y" questions at 56.7%, pricing pages led cost questions at 53.4%, and product pages took 37.7% of ChatGPT's citations. That last figure is the one clients tend to doubt: your own pages are in play, not just third-party roundups.
Run the technical checks in the same fortnight, because they can invalidate everything downstream. Search Engine Land's Olivier de Segonzac reported on 17 August 2026 that ChatGPT's crawler does not execute JavaScript, that pages over 4 MB are rejected outright, and that in instant mode the model sees roughly 200 characters of body text anchored on the H1.
Then check access. OpenAI's publisher FAQ states that a site must not block OAI-SearchBot to be included in ChatGPT's summaries and snippets, and that GPTBot is the separate training crawler. Perplexity's documentation says allowing PerplexityBot is what keeps a site in its search results, while Perplexity-User, which fires when a person asks a question, generally ignores robots.txt.
I have seen a blocked user agent in robots.txt account for a company's entire absence. Check it before you write anything.
Days 31 to 90: Source Engineering
The third process is Source Engineering: getting you onto the pages these engines actually read, then verifying the change against the tracked questions rather than against a general visibility score.
Work in the order the gap analysis gave you. Usually that means fixing or building your own comparison and pricing pages first, because you control them and they map to the highest-intent questions, then working on the third-party sources that keep citing competitors.
Re-test weekly with the same prompt set. Weekly is not a service flourish, it is the only way to tell a real change from the variance SparkToro measured.
Track the referral side too. OpenAI's FAQ confirms ChatGPT appends utm_source=chatgpt.com to referral URLs, so the traffic that does click through is separable in analytics from day one.
By day 90 you should be able to say which questions moved, which did not, and which cited sources changed hands. If you want the full sequence with the weekly cadence attached, that is the 90-Day Recommendation Sprint.
How to measure a GEO strategy without lying to yourself
There is no fixed ranking inside ChatGPT. Nobody, me included, can guarantee that a company will appear in an AI answer, and any agency quoting a "#1 ChatGPT ranking" is describing something that does not exist.
What you can report honestly is a share: out of this many runs of this defined question set, you were named in this many. Report the prompt set alongside the number, or the number means nothing.
Two other habits keep the reporting clean. Baseline before you change anything, and keep engines separate, because a Perplexity gain and a ChatGPT loss cancel out into a flat line that hides both. I wrote more about what to demand from any vendor in how to choose a GEO agency.
Where an AEO strategy fits
An AEO strategy, answer engine optimization, focuses on formatting content so an engine can lift a direct answer out of it: clear questions as headings, the answer immediately underneath, no burying the conclusion.
In practice it is a subset of the same work rather than a competing discipline. I keep the label mostly because buyers search for it, and because the formatting discipline it names is genuinely useful on comparison and pricing pages.
One question to start with
Pull your last ten sales calls and write down the questions those buyers said they had already asked an AI tool before they spoke to you. How many of them could you answer today with a page you already own?
Sources
- What ChatGPT pulls, what it shows, what it cites, July 2026, updated August 2026 (vendor research) — RESONEO — https://think.resoneo.com/chatgpt-retrieval/
- NEW Research: AIs are highly inconsistent when recommending brands or products, Jan 28, 2026 (vendor research) — SparkToro (Rand Fishkin) — https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/
- Inside ChatGPT's retrieval stack: The index, cache, and pages it actually reads, Aug 17, 2026 — Search Engine Land (Olivier de Segonzac) — https://searchengineland.com/chatgpt-retrieval-stack-index-cache-pages-485036
- Publishers and Developers FAQ, accessed Sep 17, 2026 — OpenAI Help Center — https://help.openai.com/en/articles/12627856-publishers-and-developers-faq
- Perplexity Crawlers documentation, accessed Sep 17, 2026 — Perplexity — https://docs.perplexity.ai/guides/bots
- Which Content Formats Win the Most AI Citations for B2B SaaS?, 40 prompts and 490 cited URLs, Aug 28, 2026 (vendor research, ours) — GeoCited — /blog/which-content-formats-win-ai-citations-b2b-saas
Frequently asked
How long before a GEO strategy shows results?
Position mapping is available in week one, since that is just measurement. Changes in who gets cited depend on whether the fix is on your own pages or on third-party sources, and third-party placements move on the other site's publishing schedule, not yours. Ninety days is long enough to read a trend across a defined question set, not long enough to call anything settled.
Do I need a GEO strategy if my SEO is already strong?
Strong SEO helps, given that about a third of ChatGPT's retrieved URLs sit on Google's first page in the RESONEO corpus. It is not sufficient, because retrieval is only the first step: in that same corpus, retrieved-only pages were cited 7% of the time against 74% for pages the model opened.
What is the difference between a GEO strategy and a list of GEO tactics?
The tactics are largely known and shared with SEO. The strategy is the order: measure the questions, find which sources decide those answers, then spend your effort on the ones that actually feed the answers you care about.
Can I run this in-house?
Yes, and the prompt discovery step is the part that benefits most from being in-house, because your sales team already knows the questions. The heavy part is repetition: enough runs, every week, across every engine, logged consistently.
How many questions should the plan cover?
Enough to cover each buying stage, from problem questions to "X vs Y" and pricing, since the cited page types differ sharply by stage in my own data. Cover fewer questions well rather than many badly, because a question you only test once tells you close to nothing.
Written by
Lourdes Paul Agilan runs GeoCited, a generative engine optimization agency. He works with B2B companies on how AI assistants describe and recommend them. He writes about that work here, and shares the experiments behind it. The tests that worked, and the tests that did not, written up the same way.