The 90-day GEO program

Ninety days is the honest unit for GEO. AI answers respond to real changes, new content, better crawlability, presence in the sources models cite, but the loop from action to visible movement runs through crawls, retrieval pools, and answer generation, which takes weeks, not days. This program structures those weeks: month one builds a baseline you can trust, month two turns diagnosis into work, month three measures what moved and locks in the operating rhythm. The program assumes nothing beyond a Promptwatch account. If you already run monitors, start at month two and treat month one as a checklist to audit your setup against.

​

Month 1 (days 1-30): establish a baseline you can trust

Everything you do later is measured against this month’s numbers, so the goal is representative data, not good news. Resist the urge to act on week-one numbers; a baseline built on three days of data mostly measures noise.

​

Week 1: set up the measurement

1. Create your project and complete onboarding. Setup analyzes your website and prefills your brand profile and competitor suggestions.
2. Audit the Brand Book: name, domain, and especially aliases, every name your brand actually goes by, since detection can only credit names it knows. See your brand profile.
3. Set your competitor list to the brands a customer would genuinely shortlist against you, no more. Competitors are the comparison set on the heatmap and the Brand Visibility chart, so a padded list buries the comparisons you actually care about. See identifying your competitors.
4. Build your monitors. The three-monitor structure is a solid default, and one market per monitor keeps your time series clean. Aim for around 15 prompts per monitor, sourced from real customer language. See finding the right prompts.
5. Select models: ChatGPT, Perplexity, and AI Overview are the core three; add others deliberately, since each model multiplies response usage. See choosing your models.

​

Week 2: connect the data sources

Prompt tracking tells you what models say. The other two data legs tell you what models read and what traffic they send:

• Sitemap or Search Console: the index of your own content, and the prerequisite for content gap analysis. See sitemap or GSC connection.
• Crawler logs: what GPTBot, ClaudeBot, PerplexityBot, and the rest actually fetch from your site. See log ingestion.
• Visitor analytics: the human click-throughs your AI visibility produces. See visitor analytics.

The Action Board adds a To Do item for each missing integration: sitemap, Google Search Console, crawler logs, and visitor analytics. Sitemap or GSC is enough to run content gap analysis, but each still gets its own setup action until you connect it.

​

Weeks 3-4: let it accumulate, and only fix setup

Read the dashboard weekly, not daily. The only changes worth making now are setup corrections, a missed alias, a wrong monitor country, a prompt that turned out to be too generic, because every setup fix this month is a distortion you don’t carry into your baseline. The common setup mistakes page is the audit list. At day 30, record your baseline: Visibility Score and Share of Voice overall and per model, self-citation rate, and your position on the competitor heatmap. These are the numbers day 90 gets compared against.

​

Month 2 (days 31-60): diagnose, then close the biggest gaps

With 30 days of data, differences are now signal. This month converts them into shipped work.

​

Diagnose where you lose

• Per model, not blended: filter the dashboard by each model. The platform where you’re weakest is usually where the cheapest wins live. See metrics overview for reading the metrics together.
• Per competitor: the competitor heatmap shows who beats you on which model, which is a sharper question than “how are we doing”.
• Per prompt: sort prompts by visibility, then open a prompt and read its responses. Who gets named instead, and which sources do those answers cite?
• Per page: run the Content Gap analysis, which scores how well your site covers each prompt’s underlying queries and turns misses into create-or-optimize recommendations rated by impact and effort.

​

Act on the top of the list

Work from the Action Board: the agent cross-references your gaps, sentiment, crawler activity, and offsite signals into a prioritized queue, so you spend the month executing rather than re-deriving priorities. For content work specifically:

• Take high-impact, low-effort content gap recommendations first, and hand them to the Content Agent, each recommendation carries a Run content agent or Optimize page button. See turning a gap into a brief.
• Fix crawlability blockers surfaced in Crawler Logs and site health: a page AI bots can’t read can’t be cited, whatever its quality. See crawlability.
• Start one offsite motion. If your category’s answers cite Reddit or YouTube heavily, that channel outweighs another blog post; the Reddit and YouTube strategy is the companion playbook.

Ship steadily rather than all at once: content published in week 5 has time to be crawled, cited, and reflected in answers before day 90; content published in week 8 may not.

​

Month 3 (days 61-90): measure what moved, keep what works

​

Track the causal chain, not just the score

Add every page you published or optimized to the Page Tracker. Its per-page view shows the chain you’re trying to complete: AI crawler visits first, then citations from your tracked prompts, then click-throughs. Crawls without citations point at content that gets read but not used; no crawls at all point back at crawlability. Meanwhile watch the same dashboard cuts you baselined: visibility and Share of Voice per model, self-citation rate, heatmap position. Attribute honestly, a lift on prompts you built content for is your work; a broad lift across everything may be a model update.

​

Prune, expand, and report

• Retire prompts that produced no decisions, add prompts for the gaps and fanout patterns you discovered, and label everything with topics and tags so next quarter’s analysis is a filter, not a project.
• At day 90, build the comparison against your day-30 baseline and share it as a report. Stakeholders don’t need every chart; they need “here’s where we were, here’s what we shipped, here’s what moved”.

​

What day 91 looks like

The program’s end state is a rhythm, not a finished project: a weekly dashboard read, the Action Board as the shared to-do list, content gaps feeding the Content Agent, and the Page Tracker proving which work paid off. Repeat the diagnose-act-measure loop quarterly, with each baseline stronger than the last.