How to build a GEO strategy: a 90-day plan with owners and deliverables
A GEO strategy that holds up is a 90-day loop: audit which questions matter and what AI assistants say today, publish pages that answer them with your own data, earn mentions on sites the engines already cite, fix technical basics in parallel, and re-check the same prompts monthly instead of chasing promised rankings.
Most GEO advice arrives as a pile of tactics with no order and nobody responsible for them. This guide turns the pile into a calendar: what to do in each of the next 12 weeks, who on your team does it, and what should exist when each phase ends. If you need the basics first, read what generative engine optimization is and come back.
The plan rests on what Google, OpenAI and the Federal Trade Commission have published, plus a check of our own. We asked ChatGPT eight versions of “how do I get my business recommended” to see what it advises and which sources it trusts for that advice.
The 90-day GEO strategy at a glance
Short version: two weeks of audit, four of content, six of off-site work that overlaps with the content, a technical track that runs the whole time, and two weeks to measure and decide what comes next. The overlaps are deliberate. You need your first data piece before you can pitch it, and you need the audit before you know who to pitch.
| Weeks | Phase | Main work | Owner | Deliverable |
|---|---|---|---|---|
| 1 to 2 | Audit | Prompt list, baseline of AI answers, crawler and index checks | Marketing lead with SEO support | Prompt sheet of 30 to 50 questions, baseline log, technical issue list |
| 3 to 6 | Content | Rewrite or build answer pages, publish one original data piece, write fair comparisons | Writer plus an in-house expert | 6 to 10 answer pages live and one data asset with its method |
| 5 to 10 | Mentions | Reviews, directories, list authors, trade press, communities | Owner or PR lead, customer success | Review requests built into every sale, profiles corrected, pitch list worked through |
| 1 to 12 | Technical | Server-rendered text, crawler access, structured data, speed | Developer | Fixes shipped and validated |
| 11 to 12 | Measure and reset | Re-run the same prompts, read Search Console, count leads | Marketing lead | Quarter report and a backlog for the next 90 days |
Vértigo Marketing plan. The counts are targets for a small or midsize business; scale them to your size.
In a five-person company, one person may hold three of those roles, and that works. What sinks GEO programs is a row with no name next to it. Usually it is the off-site row, because it means asking other people for things.
Weeks 1 and 2: audit the questions, the answers and the crawlers
The audit settles three things: which prompts are worth winning, what the engines say about your market today, and whether anything stops them from reading your site. Skip it and you will polish pages nobody asks about.
Build the prompt list
Collect the questions buyers ask before they pay. The best sources are already inside the company: sales call notes, support tickets, quote requests and the long, question-shaped queries in Search Console. Add the “People also ask” boxes for your main services and a few Reddit threads where people compare providers in your category.
- Category prompts: “best [service] for [type of customer] in [city]”.
- Problem prompts: “how do I fix [problem]” or “why does [thing] keep happening”.
- Comparison prompts: “[your brand] vs [competitor]” and “alternatives to [competitor]”.
- Cost prompts: “how much does [service] cost in [city]”.
- Trust prompts: “is [your brand] legit” or “[your brand] reviews”.
Aim for 30 to 50 prompts and tag each with the revenue it could bring. When we put the same planning question to ChatGPT, three of its eight answers landed on that range too. Fewer prompts leave you guessing; many more create noise you will not have time to read.
Record what the engines say today
Run every prompt in ChatGPT, Google (AI Overviews and AI Mode), Perplexity and Gemini while logged out, and paste the answers into a sheet. For each one, note whether you are named, whether your pages are cited, which competitors appear and which third-party domains the engine leans on. Then run the whole set a second time, because answers shift between runs and one pass will mislead you.
The list of cited domains is the most valuable output of the audit. It tells you which review sites, directories, lists and publications the engines trust in your market, and it becomes the target list for weeks 5 to 10.
Check the plumbing
- robots.txt and the firewall. OpenAI says sites that block OAI-SearchBot will not be shown in ChatGPT search answers, and that robots.txt changes take about 24 hours to register. Check PerplexityBot, Bingbot and Googlebot as well, and ask whoever runs your CDN whether a bot-protection rule is turning them away.
- The Search Console setting. Since August 31, 2026, every site has a Search generative AI control under Settings. Including the site is the default. A site set to exclude gets no traffic or impressions from AI Overviews, AI Mode or the AI features in Discover, so confirm nobody flipped it.
- Index and snippet eligibility. Google says a page must be indexed and eligible to show a snippet before it can appear in its generative AI features. A forgotten noindex or nosnippet tag on a key page ends the matter there.
For a quick first pass, the free GEO audit from Slibur.com runs 18 automated checks covering crawler access, llms.txt, text visible without JavaScript and structured data. It does not replace reading the answers yourself.
Weeks 3 to 6: pages that answer, data only you have, fair comparisons
Start with what you already have. Map each high-value prompt to one existing page, and create a new page only when nothing on the site could reasonably answer it. Most sites discover that their service pages describe the company and never answer the question the buyer typed.
Turn service pages into answer pages
Each priority page should answer its prompt in the opening lines, then give the conditions that change the answer, then the detail: a table, steps, prices with a date, examples from real jobs. Name the person who wrote or reviewed it. Our AEO guide has an answer-block template you can copy; what matters here is pace, and two or three pages a week is realistic for a small team.
Publish one piece of data nobody else has
Google’s guide to its generative AI features, last updated July 10, 2026, says unique, useful content will likely shape your presence in AI search more than any other suggestion in it. It contrasts commodity articles built on common knowledge with expert takes that only someone with experience could write. For a business, the cheapest way to produce the second kind is to publish your own numbers.
- A roofer: replacement cost by roof size, taken from last year’s invoices.
- A software company: how long onboarding really takes, measured across customers.
- A dental practice: the questions patients ask most before an implant consultation.
- An online store: return rates by product type and the reasons buyers give.
Put the method and the date next to the figures. A number with a method is something a journalist, a list author or an AI engine can cite. Without one, it is just a claim.
Write comparisons you would show a competitor
Buyers ask engines to compare, so give them a fair comparison to quote. Cover “[your brand] vs [competitor]”, “alternatives to [competitor]” and “best [category] for [use case]”, and say plainly where you are not the right fit. When ChatGPT laid out its own plan for us, it recommended these same pages and asked that they be accurate, fair and open about the criteria.
What you should not build is a page for every phrasing of a question. Google’s guide says that creating separate content for each variation or fan-out query, mainly to sway rankings or AI answers, violates its scaled content abuse policy. Our explainer on query fan-out shows how to cover those follow-up searches inside the pages you already have.
Vértigo Marketing runs GEO for companies in the US. Get a free review
Weeks 5 to 10: earn mentions where the engines already look
AI answers lean heavily on what other sites say about you, so the second half of the quarter goes to being present, accurately, on the domains your audit surfaced. It is slower than editing your own pages. It is also where most of the effect appears to come from.
The best public data point is an Ahrefs study of 75,000 brands, first published in May 2025 and updated in April 2026. Brand mentions across the web had a correlation of 0.664 with how often a brand appeared in AI Overviews, against 0.218 for backlinks. Brands in the top quarter for web mentions averaged 169 AI Overview mentions, while the next quarter down averaged 14. Ahrefs stresses that correlation is not causation, and the study covers Google only, but it points the same way as what the engines cite in practice.
- Reviews
- Ask every customer, every time, on the platforms the engines cited in your audit: Google Business Profile for local services, G2 or Capterra for software, Yelp, Trustpilot or an industry site elsewhere. Build the request into the end of each job so it never depends on someone remembering.
- Directories and profiles
- Claim and complete the industry directories, association listings, Bing Places and chamber pages that showed up in the answers. Keep your name, address, phone number and list of services identical everywhere, because engines cross-check them.
- Lists and roundups
- In the buying-question check from our GEO explainer, 37 of the 65 pages ChatGPT cited were lists. Write to the authors of the lists the engines quoted in your market, with accurate, current facts and your new data piece. If a list charges for inclusion, the placement has to be disclosed.
- Press and trade media
- Pitch the data you published in weeks 3 to 6 to trade publications and local outlets. A figure with a method gets picked up far more often than a company announcement does.
- Communities
- Answer questions in the forums and subreddits where your buyers compare options, as a named person who says where they work. Our Reddit SEO guide covers how to take part without getting removed.
Track this like a sales pipeline: who you contacted, when, and what came of it. At the week 11 re-check you will want to know which new mention might explain a change in the answers.
In parallel: the technical work that keeps you eligible
Nothing on this list wins a mention by itself, but any item on it can lose you one. Hand it to a developer in week 1 and review progress every two weeks.
| Item | Why it matters | How to check |
|---|---|---|
| Main text in the HTML, not only in JavaScript | Vercel found in December 2024 that none of the major AI crawlers, OpenAI’s and Anthropic’s among them, executed JavaScript; Gemini renders through Googlebot | View the page source or load it with JavaScript off; the main content should still be there |
| Crawler access | Blocking OAI-SearchBot removes you from ChatGPT search answers | robots.txt, CDN bot rules and server logs |
| Indexing and snippets | Google requires both for AI Overviews and AI Mode | URL Inspection in Search Console; look for noindex and nosnippet |
| Structured data | Not required for AI features, says Google, but it still earns rich results and labels your business, products and reviews | Rich Results Test and the Schema Markup Validator |
| Speed and page experience | Google lists low latency and a clearly separated main content area among its page experience points | Core Web Vitals report and PageSpeed Insights |
| Duplicate URLs | Google suggests reducing duplicates so crawling is not wasted on pages you do not care about | Canonical tags, parameter handling and the Pages report |
Sources: Vercel and MERJ (December 17, 2024), OpenAI crawler documentation, Google Search Central (July 2026).
Two items people expect here are missing on purpose. Google says its Search ignores llms.txt files and needs no special AI markup. Creating an llms.txt for other services does no harm, but it does not belong in week 1.
Weeks 11 and 12: measure without promising positions
There is no position one in ChatGPT. Answers change between runs, users and days, so measure shares and trends on a fixed set of prompts and treat any single answer as an anecdote.
- Mention rate: the share of your prompts in which an engine names you.
- Citation rate: the share in which it links to one of your pages.
- Source share: which domains get cited when you do not, compared with the week 2 baseline.
- Accuracy: wrong prices, old addresses, services you dropped years ago. Fix them where the engine found them.
- Google impressions: the generative AI performance report in Search Console, launched June 3, 2026 and open to all sites since August 31, shows impressions in AI Overviews, AI Mode and Discover by page, country, device and date.
- Visits and leads: sessions from chatgpt.com, perplexity.ai and gemini.google.com in analytics, plus an “AI assistant” option in the “how did you hear about us” field of your forms.
Compare like with like: same prompts, same engines, same logged-out conditions, several runs, reported as ranges. Going from 4 of 40 prompts to 9 of 40 over a quarter is a result you can defend in a meeting. “We rank first in ChatGPT” is not.
Google adds a warning worth taping to the wall. Be wary of tools that promise ranking success or claim to use internal Google metrics, since no outside tool can see its ranking or AI systems. Trackers are still handy for sampling at scale: our roundup of the best GEO tools compares prices and engines, and Slibur.com’s free AI visibility checker gives a first reading at no cost.
Decide the next 90 days
Close the quarter with three lists: what moved the numbers and gets more effort, what did nothing and stops, and which prompts are still lost and why. The second quarter usually tilts from fixing pages toward off-site work, because the on-site fixes are done and mentions take longer to build. If you would rather have a team run that phase, our GEO service for US companies starts from this same audit.
What not to do: spam, mass content and fake mentions
Every GEO shortcut has a name in a policy somewhere. These are the ones that can cost far more than they could ever bring in.
- Mass-produced pages
- Google’s spam policies, last updated August 28, 2026, describe scaled content abuse as generating many pages mainly to manipulate rankings rather than help users, however the pages are made. Programmatic pages for every city, service and question variation, written by AI with nothing added, are the textbook case.
- Fake or bought reviews
- This one is federal law, not just a search guideline. The FTC rule on reviews and testimonials, announced on August 14, 2024 and in force since October 21, 2024, bans fake reviews (AI-written ones included), buying them, rewards conditioned on a positive or negative opinion, undisclosed reviews from managers and staff, and suppressing negative ones. Knowing violations can bring civil penalties.
- Planted mentions
- Sock-puppet forum posts and fake “customer” comments are risky and weak at once. Google’s guide says chasing inauthentic mentions is less helpful than it looks, because its AI features depend on the same systems that reward quality and block spam.
- Renting someone else’s reputation
- Paying a high-authority site to host pages about you, so they rank on its strength rather than their own, is what Google calls site reputation abuse. Its November 2024 update says the policy applies even when the host site is involved in the content.
- Hidden instructions for AI
- White text telling language models to recommend you is hidden text, which Google’s spam policies have banned for years. Readers who find it tend to share screenshots.
- Contracts that promise placements
- Nobody controls what a model answers. Our breakdown of what GEO costs lists this and other red flags to look for in agency proposals.
We asked ChatGPT how to get recommended: its advice and its sources
On October 27, 2026, we put eight GEO questions to ChatGPT from a US location, phrased the way a business owner would type them. Every answer triggered a web search, and the model was reported as GPT-6. The advice was sensible and cautious, and it came almost entirely from the companies that run the engines.
| Question | Core advice | Sources it cited |
|---|---|---|
| How do I get my business recommended by ChatGPT | Complete the Google Business Profile, make the site easy to read, collect genuine reviews; no guarantee | OpenAI help center and crawler documentation |
| How can a small business show up in AI search results | Consistent business details, pages for real customer questions, local profiles | Google Search Central, Google Business, Bing Webmaster Tools |
| What is the best generative engine optimization strategy for 2026 | Seven pillars from answer content to measurement; begin with 30 to 50 buyer questions | Google Search Central, two arXiv papers, a Princeton page |
| How long does it take for GEO to work | A few weeks to several months; give it at least 90 days | Two marketing company blogs and an arXiv paper |
| How do I get my company cited in Google AI Overviews | Target real buyer questions, answer them on the page, stay indexable | Google Search Central and Search Console help |
| Step by step plan to improve my brand’s visibility in AI answers | A 90-day plan: audit, foundation, content, outside authority, measurement | Google Search Central, once through a proxy copy |
| What should I avoid when optimizing for AI search | Ten mistakes, from AI content at scale to bought mentions and guaranteed placements | Google Search Central only |
| How do I measure if ChatGPT mentions my brand | Mention rate, share of voice, recommendation rate, citations, trend over time | Tool vendors: Ahrefs, Conductor, Measure |
Own check: 8 questions to ChatGPT from the US, October 27, 2026 (data/blog/pruebas/geo-strategy.json).
Our own check, run on October 27, 2026. AI answers change over time, so the same question may get a different answer today.
Three patterns stood out. Seven of the eight answers said, one way or another, that nobody can guarantee a mention. Google’s developer documentation was cited in five of the eight, and ten of the twelve background searches ChatGPT ran included words like “official”, “documentation” or “Google Search Central”. Agency blogs made it into a single answer, the one about timelines.
A fourth pattern was less tidy. Among the pages ChatGPT retrieved were copies of Google and OpenAI documentation on proxy and mirror domains, and one answer cited a proxy copy instead of Google’s own page. Retrieval is good but not flawless, one more reason to keep your facts identical everywhere they appear.
We also ran the first question through Google. Its AI Overview cited two agency articles and two YouTube videos, and it stated as fact that conflicting business details make AI models skip a business, without a source we could trace. Consistent details are good practice anyway. Just know that a lot of GEO advice in circulation is a confident guess.
So the lesson is not ChatGPT’s advice, which mostly mirrors Google’s guide. It is that for questions about itself, the engine reached for official sources. For questions about your market it will reach for whichever sources it trusts there, and the week 2 audit is how you find out which ones they are.
Frequently asked questions
What is a GEO strategy?
A GEO strategy is a plan for getting your business named and cited in AI answers from ChatGPT, Google AI Overviews and AI Mode, Perplexity and Gemini. In practice it combines four workstreams (answer pages, original data, third-party mentions and technical access) with a fixed set of prompts that measures progress.
How long does GEO take to work?
Technical fixes can register within days; OpenAI says robots.txt changes take about 24 hours. Content and mentions take longer, usually a quarter or more before the share of answers that name you moves in a way you can trust. When we asked, ChatGPT itself suggested giving it at least 90 days.
What are GEO best practices in 2026?
Answer real buyer questions on your own pages, publish data nobody else has, get reviewed and listed on the sites engines already cite, keep content readable without JavaScript and open to AI search crawlers, and measure a fixed prompt set over time. Avoid mass-produced pages, fake reviews and planted mentions.
How do I do GEO with a small team?
Keep the phases and shrink the counts: 20 to 30 prompts, two answer pages a week, one data piece per quarter and a review request after every job. One person can coordinate it if a developer clears the technical list at the start.
Is a GEO strategy different from an SEO strategy?
Less than the labels suggest. Google says optimizing for its AI features is still SEO, and most of the work overlaps. GEO puts more weight on off-site mentions and on measuring answers rather than rankings, and it has to account for engines like ChatGPT that run their own crawlers.
Do I need llms.txt or schema markup for GEO?
Neither is required for Google. It says its Search ignores llms.txt and that no special schema exists for AI features, though structured data still helps with rich results. Some other services may read llms.txt, so adding one is harmless, just not urgent.
How do I know if my GEO strategy is working?
Compare the same prompts on the same engines against your week 2 baseline: more answers that name you, more that cite your pages, fewer errors about you. Add the generative AI impressions in Search Console, visits from AI assistants in analytics and leads who say they found you through one.
Keep reading
- What is generative engine optimization? The study, the numbers and first steps
- Query fan-out: the hidden searches behind AI answers, and how to match them
- How much does GEO cost? What agencies, freelancers and in-house teams charge in 2026
- Entity SEO for LLMs: how to make language models know, find and cite your brand
- GEO agency services in the United States
- AEO and GEO tools, compared
Sources
Figures and claims in this guide come from these sources:
- Google Search Central: Optimizing your website for generative AI features (updated Jul 10, 2026)
- Google Search Central: Spam policies for Google web search (updated Aug 28, 2026)
- Google Search Central blog: Search generative AI performance reports (Jun 3, 2026)
- Search Console Help: Search generative AI control
- Google Search Central blog: Updating our site reputation abuse policy (Nov 19, 2024)
- OpenAI: Overview of OpenAI crawlers
- Vercel and MERJ: The rise of the AI crawler (Dec 17, 2024)
- Ahrefs: AI Overview brand visibility factors, 75K brands (updated Apr 27, 2026)
- FTC: Final rule banning fake reviews and testimonials (Aug 14, 2024)
By the Vértigo Marketing Published Reviewed