How to Do GEO: 7 Steps to Get Cited by ChatGPT and Perplexity
A practical 7-step GEO process: audit your AI visibility, fix your entity, restructure content answer-first, win the cited sources, and measure share of voice monthly.
Doing GEO comes down to a repeatable seven-step loop: audit your current AI visibility, fix your entity footprint, restructure key pages answer-first, get into the sources AI already cites, publish original data, add structured markup, and re-measure monthly. Structural fixes typically show citation movement within 30 to 60 days; authority building takes three to six months.
That's the whole system. The rest of this guide is how to run each step properly, with the templates to copy. If you're still deciding whether GEO deserves the effort, our complete guide to generative engine optimization covers the why; the short version is that AI-referred visitors convert at several times the rate of traditional organic traffic, and roughly half of brands still have no GEO strategy at all. The window is open and it will not stay open.
Step 1: Audit your current AI visibility
Before optimizing anything, find out where you stand. Run 20 buying-intent prompts across ChatGPT, Perplexity, and Gemini, and log every brand each engine names.
Build the prompt list from real buyer language, not keywords. Copy this template and fill in your category:
- best [category] for [your core customer]
- best [category] for [adjacent customer]
- [category] with [your differentiating feature]
- [competitor] alternatives
- [competitor] vs [competitor]
- is [your brand] good / legit
- how much does [category] cost
- [category] for beginners
- cheapest [category] worth using
- [category] that [solves your #1 use case] Run each in a fresh chat with search enabled, record which brands appear, in what order, and with what framing. Also expand the sources panel and note which pages the answer was built from. Those cited pages become your Step 4 target list, and the whole exercise becomes your baseline: your share of voice is simply the percentage of prompts that name you.
Do this manually the first time even if you plan to buy tooling. Seeing the raw answers teaches you how your category is actually discussed, which no dashboard fully conveys. (When you do want it automated, we compared the options in our best GEO tools guide.)
Step 2: Fix your entity footprint
AI engines hedge on brands they can't confidently identify, so the second step is making every description of your company agree. Your name, one-line description, category, and pricing should match across your website, LinkedIn, Crunchbase, G2 or the review site for your category, and every directory you're listed in.
The test is simple: ask each engine "what is [your brand]?" If the answer is vague, outdated, or describes you as the wrong kind of product, your entity is weak. Pick the exact category label you want to win, write one canonical sentence describing what you do, and repeat both everywhere, verbatim. Consistency is the signal; creative variation between profiles actively hurts you here.
This is also where you check the technical floor. Fetch your key pages with GPTBot's user agent and confirm real content comes back rather than an empty JavaScript shell, and check that your robots.txt allows the search crawlers (OAI-SearchBot for ChatGPT, PerplexityBot) even if you block training bots. An engine can't cite what it can't read, and this single check invalidates everything else if it fails.
Step 3: Restructure your key pages answer-first
AI systems extract direct-answer content disproportionately, so rewrite your most important pages to answer first and elaborate second. In one Search Engine Land audit of high-traffic domains, a concise self-contained answer placed directly under a question-format heading was the single strongest predictor of ChatGPT citations.
The pattern to apply on every page that matters:
Question-format H2s ("What does X cost?" instead of "Pricing"). Directly under each H2, a two-to-three sentence answer capsule that would make sense quoted out of context. Then the depth. Each section self-contained, because retrieval pulls passages, not pages.
One more structural reality to design for: when someone asks an AI a complex question, the engine breaks it into three to five sub-queries and searches each independently before synthesizing. "Best CRM for a 20-person remote team" fans out into feature, pricing, team-size, and review sub-queries. The brands that get cited are the ones whose content surfaces for the sub-queries too, which is the practical argument for covering your topic as a cluster rather than one mega-page.
Start with your three highest-value pages, not your whole site. Homepage, main product page, and your best-performing article.
Step 4: Get into the sources AI already cites
Your own site is rarely the deciding source for recommendation prompts, so step four is earning presence in the pages the engines retrieve. You collected this target list in Step 1's sources panel; now work it.
The pattern across engines is consistent: listicles and comparison articles carry a large share of recommendation citations, ChatGPT leans heavily on Wikipedia and established publications, Perplexity over-weights Reddit, and Google's AI Overviews cite Reddit and YouTube more than any brand site. So the work splits three ways: pitch the specific "best X" articles already being cited for inclusion, build a genuine presence in your category's subreddits, and keep your review-platform profiles active and honest.
This is the slowest step and the most defensible one. We've written the full playbook, including what gets brands banned, in how to get your brand recommended by ChatGPT.
Step 5: Publish original data
Statistics are the single most citable asset you can produce. The Princeton GEO research found that adding quantitative statistics lifted a source's AI visibility by up to 40%, with attributed expert quotes close behind, while keyword stuffing did nothing or backfired.
You don't need a research department. One data post per quarter from what you already have: aggregate patterns from your product usage, a 50-person survey of your customers, an original analysis of public data in your category. Package the findings as specific, attributable numbers ("47% of X do Y, according to [brand]'s 2026 analysis of Z"), because every article that repeats the stat becomes another source pointing the engines at you.
Step 6: Add structured markup
Schema won't earn citations on its own, but it makes your content unambiguous to machines, which is the point. Ahrefs' controlled testing found schema alone didn't directly lift AI citations, so treat it as the reliability layer rather than a growth lever: it ensures that when an engine reads your page, it correctly identifies what you are, who wrote it, and which block answers which question.
Priority order: Organization schema (feeds your entity from Step 2), FAQ schema on every article, Article schema with real author credentials, and HowTo schema on process content so each step is independently extractable.
Step 7: Re-measure monthly and iterate
GEO is a loop, not a project. On the first of every month, re-run the exact Step 1 prompt list in fresh chats and log three numbers: share of voice (what percentage of prompts name you), citation sources (which of your pages, and which third-party pages, the answers were built from), and sentiment (how you're framed when you do appear). Add an AI-referral segment in your analytics to track visitors arriving from chatgpt.com, perplexity.ai, and gemini.google.com.
Expect structural work (Steps 2, 3, 6) to move numbers within 30 to 60 days, and authority work (Steps 4, 5) to compound over three to six months. When a competitor appears in a prompt where you don't, pull up their cited page and compare it to yours against the Step 3 structure; the citation gap is usually visible in the first hundred words.
Running the loop on autopilot
Every step above is manual-doable, and the honest cost is 10 to 15 hours a week of consistent execution: the monthly prompt audits, the answer-first rewrites, the Reddit presence, the content production. This is exactly the loop Okara's agents run daily. The GEO agent tracks your visibility and sentiment across ChatGPT, Perplexity, Gemini, and Claude and scores you against competitors. The SEO and coding agents surface the structural fixes with copy-paste snippets, two per day. The writer agent drafts the answer-first content targeting the prompts you're losing, and the Reddit agent finds the threads where your buyers are asking. Draft-first, $99/month flat, and your first visibility audit takes minutes at okara.ai.
FAQ
How long until ChatGPT mentions my brand? Structural and entity fixes can show movement in 30 to 60 days, and in tracked accounts the first new citations after entity and schema work have appeared in as little as two to three weeks. Building the third-party authority that wins recommendation prompts takes three to six months.
How many prompts should I track? Twenty is the practical minimum for a stable share-of-voice number: ten recommendation prompts, five comparison prompts, five brand and pricing prompts. Re-run the identical list monthly, because engines vary their answers and consistency in your measurement is what makes the trend real.
Does posting on Reddit really help GEO? Yes, and it's among the strongest signals for recommendation queries, because engines treat community consensus as trust. It only works done honestly: useful answers, disclosure that you built the product, mentions only where they're genuinely the answer. Astroturfing gets banned by moderators and poisons the exact source you need.
Can I do GEO without doing SEO? The foundations overlap too much to separate. Crawlable pages, quality content, and authority feed both, and well-ranked pages remain a major citation source. Run GEO as a layer on top of SEO rather than a replacement.
What's the single highest-impact step? For most sites, Step 3. Restructuring your key pages answer-first is fully in your control, costs nothing, and targets the strongest known citation predictor. Do it the same week as your Step 1 audit.