Get cited by AI search

Built from the system Jason Patel, CEO and founder at Open Forge AI, described on GTM Vault GTM 34 (https://pro.gtmvault.co/episodes/gtm-34). Published free by GTM Vault. Numbers below are the operator's own, from the episode; your results will differ.

When to use this

Use it when buyers in your market already ask ChatGPT, Perplexity or Claude what to buy and you do not know whether your company is in the answer. Patel places the right moment after the founder-led sales stage, once a marketer is scaling marketing, at a couple million dollars in ARR or higher (GTM 34, 25:34 and 26:35). Before product-market fit, he says to focus on product and revenue instead (GTM 34, 26:35).

The system on the record

Patel frames generative engine optimization as a change in how people search. A chatbot returns one to five answers instead of 10 blue links, and the goal is to be one of them (GTM 34, 1:09). Models place a brand as an entity on a knowledge graph, associate topics with it, and judge whether it is trustworthy and whether its content answers the question cleanly (GTM 34, 2:17 and 3:27). Because "AI is probabilistic," visibility means being seen, cited and selected as often as possible, not holding a fixed rank (GTM 34, 5:38 and 6:46).

His method is a three-legged stool (GTM 34, 6:46 to 9:44):

1. Commercial and transactional content. Pricing charts, comparison charts and posts on your product versus the competition, for buyers already primed to buy. Broad "what is" content risks being summarized into a generic answer with no link or citation (GTM 34, 16:39).
2. Third-party citations. Relevant links matter more than many links: top 10 articles, industry roundups, publications in your space, and B2B review sites such as G2 and Capterra (GTM 34, 7:55). In rapid fire he names the number of third-party citations as the most underrated authority signal (GTM 34, 41:42), and says unlinked brand mentions also build trust (GTM 34, 29:45).
3. Structured data. Schema tells the crawler whether a page is a product, FAQ, blog or landing page and what is on it, such as a pricing table or image. CMS plugins make this a non-technical job (GTM 34, 7:55 and 8:49).

On page structure he adds: "No paragraphs over four sentences long." Use bullets, make each H2 a question or a sentence someone would say, and make the first one or two sentences under it the answer, because models lift snippets, not pages (GTM 34, 20:59 and 22:10). To check indexability, set up Google Search Console: if Google can crawl you, he expects ChatGPT can too (GTM 34, 22:10 and 23:17).

Two loops close the system. AI traffic is rented, so give arriving visitors a clear reason to hand over an email and cultivate them through a newsletter and YouTube (GTM 34, 12:03 and 23:17). Attribution breaks because you cannot see what the buyer typed, so ask how they heard about you, what their pain points are and which competitors they considered, then build content from the answers (GTM 34, 13:10, 14:17 and 17:50). For agents, be explicit about ICP and pricing and compare honestly against competitors (GTM 34, 35:34 and 37:30).

The result

1. Timing: Patel has seen someone show up 24 hours later, and generally says to give it 3 weeks to a month, with visibility ROI possible after a couple of weeks of doing GEO correctly (GTM 34, 26:35).
2. Team size: a head of GEO can be one or two people, with a junior team member for citation outreach (GTM 34, 28:43).
3. Traffic quality: he would rather have 25 visitors primed to buy from ChatGPT than a thousand tire kickers from traditional Google search (GTM 34, 30:31).

Run it on your company

1. Ask the user once, in one message, for: (a) company name, domain and one-line value proposition; (b) ICP and two or three buyer personas; (c) three to five named competitors; (d) current ARR stage and whether a marketer owns growth; (e) CMS and whether Google Search Console is set up; (f) URLs of the pricing page, comparison pages and top 10 commercial pages; (g) existing review profiles and any roundups that already mention them; (h) any customer answers to "how did you hear about us" or sales call notes.
2. Run the readiness check. If they are pre-product-market fit, say so and stop at a short note. If Search Console is not set up or shows crawl errors, make fixing that the first task.
3. Build the query panel. Write 25 queries the ICP would type into a chatbot, tagged Commercial, Transactional or Comparison, with at least five naming a competitor ("X vs Y", "alternatives to X", "best tool for [job] for [persona]"). Exclude broad "what is" queries. Output a table with columns Query, Persona, Intent.
4. Take the recall baseline. Because answers vary, run each query three times in each of ChatGPT, Perplexity and Claude, or give the user the exact protocol if Claude cannot run them directly. Record for each run: brand named (yes or no), brand linked (yes or no), competitors named. Output a CSV and a summary table with Recall % per engine (runs where the brand is named divided by runs) and Citation % (runs where it is linked).
5. Write the page plan. For each commercial page and each missing page the panel exposes (pricing, comparison, alternatives, implementation FAQ), give: target queries, a question-style H2 outline, the liftable first sentence under each H2, paragraph limit of four sentences, and the schema type to add. Output one Markdown table per page.
6. Build the citation target list. List 20 third-party properties: review sites, roundups and top 10 articles that already appear in the baseline answers, category publications and newsletters. Columns: Property, Type, Appears in baseline (yes or no), Competitors listed, Ask, Owner.
7. Run the agent clarity audit. Score the site yes or no on: ICP stated, pricing visible, honest comparison against named competitors, key facts in HTML rather than images. Output the gaps as a fix list.
8. Set measurement. Re-run the panel monthly and track recall query by query against each competitor, on the queries that matter. Add a "how did you hear about us" field naming AI assistants as an option, and track conversion and time on page from AI-referred visits rather than traffic, which fits into an existing SEO or Looker dashboard (GTM 34, 31:26 and 33:14). As Patel puts it: "Forget traffic."

Tools on the record

1. ChatGPT: generative engine optimised for.
2. Perplexity: AI search surface tracked.
3. Claude: assistant surface where visibility is tracked.
4. G2: review source seeded so engines cite the brand.
5. Capterra: review source seeded for citation.
6. Google Search Console: first setup step for GEO measurement.

Any tool doing the same job works; the method is the point. Full stack: https://pro.gtmvault.co/stack/operator/jason-patel

Where it breaks

1. Too early. Without product-market fit and a marketer scaling demand, Patel says GEO is not the priority (GTM 34, 26:35).
2. Crawl problems. If Google cannot index the site, the rest of the plan has nothing to stand on (GTM 34, 23:17).
3. Generic content. Broad informational posts get summarized without citation, which he calls bad ROI (GTM 34, 16:39 and 17:50).
4. Gaming the engines. Patel warns that bad-faith attempts to game AI agents lower trust for everyone; dishonest comparison pages are the same mistake (GTM 34, 39:46).
5. A moving target. He notes that how models surface sources may change by the time you act, and that much of it is a black box (GTM 34, 3:27 and 9:44), so treat the monthly re-run as the source of truth.