A 30-day GEO starter plan

Days 1 to 7: baseline prompts and first runs

The first week buys one thing: a baseline you trust. Write 15 to 25 buyer questions, start daily runs, and record where you stand before changing anything on your site.

1. Draft the prompts. Category questions (“best X for small teams”), comparison questions (“X vs Y”), pricing questions, and three or four about your brand directly. The setup wizard's prompt suggestions are a reasonable starting list if you would rather not begin from a blank page.
2. Start daily runs. In advisible every prompt runs each day on ChatGPT, Perplexity, Gemini, Grok and Claude through your own provider key, and Run now covers the day you want an answer immediately.
3. Read the first answers in full. Not the dashboard yet. Open the answer archive and read what the engines actually say: which brands, which sources, which claims are wrong about you.
4. Write down the starting numbers once a few days of runs exist: Visibility per prompt, each with its “based on N answers” line, plus the list of domains appearing in your sources. This is the page you compare day 30 against.

Days 8 to 14: canonical facts and the worst stale page

Week two fixes what you control completely: the facts your own site states, and the single worst stale page feeding wrong answers.

1. List the canonical facts. Price, plan names, who the product is for, integrations, support hours, company basics. Each should be true in one place and identical everywhere it appears: pricing page, docs, help center, directory profiles.
2. Register them as brand facts. In advisible, each answer then gets a verdict of Contradicted, Supported, Inconclusive or Not covered under Brand Perception in the Fact-checking tab, which is how you will know in week four whether the fixes landed.
3. Find the worst stale page. Look at the sources behind the wrong claims from week one and search your own domain for that claim. Old pricing, a retired feature, a comparison written before your last launch: that page is usually the culprit and the fastest fix.
4. Update it properly. Current facts in the first lines of the affected section, a date that is true, and the corrected URL kept in your sitemap.
5. Check crawl access once. robots.txt should allow OAI-SearchBot, PerplexityBot and Claude-SearchBot, or none of the rest matters (see how engines choose sources).

Days 15 to 21: the source pages answers already cite

Week three happens mostly off your own domain: the two or three review, profile or doc pages that your category's answers actually pull from.

1. Export your source list. The Sources view with CSV export gives you the domains feeding answers about you, with counts. Rank the third-party ones.
2. Pick the two or three you can act on. A review profile you have never claimed, a directory listing with last year's pricing, a comparison that names you wrongly, an active forum thread where the complaint lives.
3. Fix the facts and reply where the rules allow it. Corrections on third-party domains travel into answers faster than another blog post on your own site, because they are the sources the engines were reading in the first place.
4. Publish one citable number. A price table, a benchmark, a spec sheet other writers can copy with your name attached.

Days 22 to 30: measurement discipline and the day-30 review

Runs stay daily; reads happen weekly, on the same day, against the same prompt set. On day 30 you compare like for like and write down what moved.

The discipline exists because answers change run to run, and a dashboard checked daily will hand you a bad day and a good day in the same week. Source mixes move fast too: Semrush watched ChatGPT's Reddit citation share fall from close to 60% of responses to around 10% within weeks in September 2025 (their three-month study), while the other engines barely moved. Weekly reading keeps you from calling that your doing.

For the review itself: Visibility per prompt with its sample size, which claims still read Contradicted in Fact-checking, which domains now appear in your sources, and how the gap list in Competitor gaps has shrunk. Four numbers, written down, next to the week-one page.

What not to do in the first 30 days

The first month is when most programmes break their own data. Restraint here is worth more than an extra twenty pages of content.

• Judge from one screenshot. A single answer proves nothing; the count over a week of runs is the first reading worth having.
• Rewrite forty pages at once. If everything changes, you will never know which change did anything. Fix a few, keep the rest stable, and note the dates.
• Chase every keyword-shaped prompt. Fifteen to twenty-five sharp questions beat two hundred sloppy ones, because each prompt costs an answer every day and your attention per prompt.
• Flood forums with links. It breaks the rules of the places buyers actually read, and those threads are the sources you were hoping to earn.
• Block the AI search bots out of caution. They are the retrieval layer; keeping them out removes you from the answers you are trying to appear in.
• Promise a percentage by day 30. Commit to the process and the baseline instead, and bring numbers when they have enough answers behind them.