Work plan

90-day GEO plan: from not knowing what AI says to improving with data

A 90-day GEO plan has five phases: measure your baseline, fix technical access, create or improve the pages that answer your key questions, earn presence in the sources AI cites, and measure the result to plan the next quarter. It can’t guarantee AI will recommend you, but it gives you data to know what works. GeoRush covers every phase.

Updated on

Two women planning in front of a wall covered in colorful sticky notes

Why 90 days

Ninety days is a reasonable window for a first full GEO (generative engine optimization) cycle. It leaves time to measure your baseline properly, fix technical issues, publish new content and let crawlers read it, and see the first effects on engines that search the web when they answer. Anything that depends on model training takes longer and probably won’t show in this first quarter.

This plan is designed for a small team: one marketing or content person leading it, occasional help from development or technical SEO and, if you have one, someone in communications. Scale it to your resources: doing 15 questions well beats doing 100 badly.

The plan at a glance

PhaseDaysGoalDeliverableControl metric
1. Baseline1–15Know where you stand.Question list, initial measurement, competitor and source map.Mention rate, share of voice and citation by engine.
2. Technical access10–30Let AI read you.Reviewed robots.txt, indexable pages, structured data, official fact sheet.Crawlers allowed; key pages indexed.
3. Content25–60Answer your key questions.5–10 new or improved pages.Coverage: % of questions with a page that answers them.
4. Sources and corrections40–80Be where AI reads.Priority source list, outreach done, errors corrected.Sources won; wrong claims corrected.
5. Measurement and report75–90Know what worked.Before/after report and next-quarter plan.Change in metrics with margins of error.

The phases overlap on purpose: you can start writing content while development fixes the technical side.

Overhead view of a diverse team gathered around a white table with laptops

Phase 1 (days 1–15): measure your baseline

  1. 1

    Write your buying questions

    Between 20 and 50 real questions, grouped into 4–6 topics: “best X for Y”, “alternatives to Z”, “X vs Y”, “how much does X cost”. If you have Google Search Console, start from real searches.

  2. 2

    Choose engines and markets

    The assistants your customers use, and their countries and languages. At minimum: ChatGPT, Gemini, Perplexity and Google AI Overviews.

  3. 3

    Measure several times

    Every question on every engine, with repetitions. Log mentions, position, tone and cited sources.

  4. 4

    Identify competitors

    Who appears when you don’t. Brands you never considered direct competitors often show up.

  5. 5

    Map the sources

    Which domains are cited most for your questions, and what type they are.

  6. 6

    Check accuracy

    Log every false or outdated claim about your brand.

Phase 2 (days 10–30): open the door to AI

  • Review robots.txt: allow the search crawlers (OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot, Bingbot) and decide deliberately about training crawlers and Google-Extended. Guide: robots.txt for AI bots.
  • Check that key pages are indexable: no noindex, no nosnippet, content in the HTML.
  • Review speed and crawl errors on important pages.
  • Create your official fact sheet and align your site, profiles and JSON-LD with it. Guide: schema for AI.
  • If you like, publish an llms.txt; note that Google says its Search doesn’t use it. Guide: what is llms.txt.
  • Remove or redirect obsolete pages with old prices or terms.

Phase 3 (days 25–60): create the pages that answer

  • Match questions to pages: which questions have no page of yours that answers them?
  • Prioritize by estimated impact: buying-intent questions where competitors appear and you don’t.
  • For each priority question, create or improve a page with the answer up top, concrete dated facts and something original. Guide: content AI cites.
  • Compare each page with the pages AI cites for that question and cover what is missing.
  • Publish pricing, comparison and “who it’s for” pages first if they don’t exist: they are usually the biggest gaps.
  • Show the last-updated date and link the new pages from existing ones.

Phase 4 (days 40–80): earn presence in the sources

  • Pick 10–20 priority sources: the ones AI cites often for your questions that don’t mention you.
  • Contact each one with useful, verifiable information, not a link request. Guide: digital PR for AI.
  • Complete and verify your profiles on relevant directories and review platforms, and ask real customers for reviews.
  • Fix the errors found in phase 1, on your site and on third-party sites. Guide: AI giving wrong information about your brand.
  • Log every contact and its status so you can tie it to the metrics later.

Phase 5 (days 75–90): measure, report and decide

  • Repeat the measurement with the same questions, engines and markets as phase 1.
  • Compare before and after with margins of error, by engine and topic. Guide: GEO metrics.
  • Tie changes to actions: which questions improved after which page or which source was won.
  • Review traffic from assistants and impressions in Search Console’s AI features report.
  • Write a short report: what changed, what didn’t, what you learned, and next quarter’s plan.

Who does what

RoleResponsibilityIndicative effort
Marketing or content leadLeads the plan, defines questions, writes or commissions content, reports.Most of the work.
Development or technical SEOrobots.txt, indexing, structured data, speed, redirects.Concentrated in phase 2.
Communications or PROutreach to sources, media, directories and reviews.Concentrated in phase 4.
LeadershipApproves the official fact sheet and reviews the final report.Occasional.

Indicative effort only: it depends on your site’s size, how many questions you track and how much content is missing.

Weekly tracking template

A simple sheet to keep the thread. One row per week:

geo-tracking.csv
week,phase,actions_done,pages_published,sources_contacted,errors_fixed,mention_rate,interval,share_of_voice,citation_rate,notes
1,baseline,"questions defined; initial measurement",0,0,0,,,,,
2,baseline / technical,"source map; robots.txt review",0,0,0,,,,,
...

How GeoRush fits the plan

GeoRush automates the measurement in phases 1 and 5 across 9 engines, with every rate shown alongside its margin of error. Prompt research helps you build the initial list from your market and Search Console. The action plan turns the data into tasks ordered by estimated impact, with steps and ready-to-paste fixes, and measures before and after each completed task. The technical audit covers phase 2, the content optimizer phase 3, and get cited phase 4.

An indicative usage estimate: 30 questions on ChatGPT, Perplexity, Gemini and Google AI Overviews cost 180 credits per run (30 × (1 + 1 + 2 + 2)). Measuring once a week comes to about 720–780 credits a month, before any extra AI actions, within the Starter plan’s 1,500. See the plans on pricing.

Common mistakes

  • Starting to write content before measuring the baseline.
  • Changing questions mid-quarter and losing the comparison.
  • Doing everything at once without logging what changed and when.
  • Giving up in week 6 because “AI hasn’t changed”: effects arrive with a delay.
  • Promising leadership guaranteed results: nobody controls AI answers.

Start phase 1 today

The free check gives you a first snapshot in a minute: three buying questions, a technical audit and the first fixes.

Sources

Official documentation and primary sources checked on November 17, 2026. These policies change often: check the source before making important decisions.

Frequently asked questions

On engines that search the web when they answer, first changes can show within weeks, once crawlers read your new pages. Anything that depends on model training takes months. These are estimates: only measurement confirms them.

Between 20 and 50 buying questions, grouped into a few topics. Tracking a few questions well beats tracking many badly, and keeping a stable core lets you compare.

Measure first, then fix any technical issue that stops AI from reading you, and start on content in parallel. Without technical access, even the best content can’t be cited.

Not necessarily: you can measure by hand by asking the questions and logging the results. A tool like GeoRush saves that work and adds margins of error, source analysis and task tracking.

You repeat the cycle with what you learned: keep the core questions, add new ones, scale what worked and rethink what didn’t. GEO is ongoing work, like SEO.

Keep reading