Your First 30-day Action Plan

A 30-day GEO action plan turns your baseline findings into five specific actions, each with one owner, one metric, a due date and a clear "done when", and ends with a re-scan on the same settings. It's the step between knowing where you stand in AI answers and changing it. Lesson 8, the last lesson of GEO 101, shows how to turn findings into actions, prioritise them, write them so they get done, run the month, and read the day-30 re-scan without fooling yourself.

In this lesson

• Where the plan starts
• From finding to action
• Prioritising: impact, effort, confidence
• Writing actions that get done
• Choosing the metric
• From MentionX data: what a realistic target looks like
• Worked example: Ledgerline's 30-day plan
• Getting the plan agreed
• Running the 30 days
• Reading the day-30 re-scan
• What if nothing changes?
• Common plan mistakes
• Cycle 2 and beyond
• Exercise
• Quiz
• Sources

Anything marked "Our view" is our opinion. Ledgerline is the fictional company used throughout this course.

Where the plan starts

Start with the three findings you wrote in Lesson 7. Each one should already have a number, evidence, a likely cause and a suggested action. If a finding has no evidence, go back and get it before you plan around it.

Two rules:

1. Every action traces to a finding. It keeps the plan honest and the re-scan readable.
2. Accuracy findings go in, whatever else does. Wrong facts about you mislead buyers who are already close to choosing you.

From finding to action

For each finding, list the possible responses, then choose the most direct: the action closest to the likely cause.

Our view: the most direct action is rarely the most exciting. A new content hub sounds better in a meeting than "add a paragraph to an existing post". The paragraph usually works faster, because engines are already citing that page.

Prioritising: impact, effort, confidence

Score each candidate from 1 to 3:

Add the three scores and take the top five. One or two quick wins (effort 3) are worth including even if their impact is modest. Early visible progress keeps the plan supported.

Our view: treat confidence honestly. An action that scores 3 on impact and 1 on confidence is a bet. Include at most one bet per cycle.

Writing actions that get done

Use the format Action, Owner, Due, Metric, Done when.

• Action: a verb and a specific object. "Rewrite the duplicate-payments guide with a product section," not "improve content."
• Owner: one named person. Others can help, but one person is accountable.
• Due: a date within the 30 days.
• Metric: the number you expect to move, and on which prompt group.
• Done when: a check anyone can verify. "Page live and recrawled," "request sent and logged."

"Done when" matters most. Without it, actions drift to 80% finished and stay there.

Choosing the metric

Match the metric to the finding:

Some actions also need an output metric that you can check within days: the page is live, the request was sent, the profile is complete. Report both. Outputs tell you the work happened. Outcomes, measured at the re-scan, tell you whether it worked.

From MentionX data: what a realistic target looks like

Method. We ran read-only aggregate queries over completed MentionX runs from 22 August to 29 October 2026 (UTC): 2,154 runs across 18 projects, mostly B2B software, engines via API with web search or grounding on. A separate variance cut covered 28 prompt-and-engine pairs repeated 643 times between 4 September and 19 October 2026. Brands are anonymised.

Three numbers help set targets:

• Answers named few tracked brands. On average, a run named 1.38 of the brands each project tracked. Answers aren't long lists of every vendor. Being one of the one or two named is the real goal.
• Placement was usually low. In unbranded answers where the tracked brand appeared in a ranked list, it was first 18% of the time and 6th or lower 56% of the time (102 answers).
• Repeat runs varied. In 11 of the 28 repeated pairs, the brand appeared in some runs and not others. In 19 of 28, the number of tracked brands named changed between runs.

How to read it. Set targets for prompt groups, not single prompts. Single prompts flip between runs without anything changing. A realistic first-cycle target is a clear move on one intent group plus actions completed, not a jump in the overall rate.

Caveats. The data covers 18 projects and uneven engine coverage. It shows variation exists, not how big it'll be for you.

Worked example: Ledgerline's 30-day plan

Illustrative example. Ledgerline, its competitors, people and numbers are invented.

From Priya's three findings (problem prompts, comparison sources, pricing accuracy) the team listed nine candidate actions and scored them. The top five:

Dropped for now: a new content hub (low confidence, high effort) and paid placements (not tied to a finding).

Targets, written before the re-scan: problem-prompt mention rate up from 7% (n=54) by at least 10 points in two consecutive scans, and branded accuracy errors at zero.

Getting the plan agreed

Owners outside marketing, such as web, product or sales, need a reason to give you their time. Show each one the finding behind their action, two example answers and what "done" looks like. Ask for a date rather than a commitment to "look at it". A plan shared with three example answers and a clear ask is usually approved in one meeting. A plan shared as a spreadsheet of percentages tends to sit in someone's inbox.

Running the 30 days

Keep a change log: the date each page went live, was submitted in a sitemap and was recrawled. Google documents sitemaps as a way to tell it about new and updated pages (Sitemaps overview). When the numbers move, or don't, the log is how you explain why.

Reading the day-30 re-scan

Use the same method note, with no new prompts or engines. Then:

1. Compare each target metric with the baseline, with n.
2. Apply the Lesson 6 tests: is the change larger than the margin, and does it show up across several prompts?
3. Mark each target moved, too early or no change.
4. Treat a single scan's move as provisional until the next scan confirms it.

Our view: report the outputs first ("5 of 5 actions done, 3 pages recrawled"), then the outcomes. It keeps the review honest when the numbers are noisy.

What if nothing changes?

It's common after 30 days, and it isn't failure. Check:

• Recrawl: were the changed pages actually fetched again? If not, improve internal links and sitemaps, then wait.
• Sample size: could your scan detect the change you hoped for? Small cells swing either way.
• Cause: did the action address the cause you found, or a neighbouring one? Reread the worst answers.
• Lag: third-party fixes (reviews, roundups) often take weeks to be published and picked up.

Then plan the next cycle with the next five. GEO builds over months.

Common plan mistakes

• Too many actions. Ten actions in 30 days usually means three get done well. Five is a ceiling, not a target.
• Actions without findings. "Add FAQ schema everywhere" because a blog post said so. Google says no special schema is needed for AI features (Google Search Central). Without a finding, you can't tell whether it helped.
• Shared owners. Two owners means each assumes the other is on it.
• Moving the goalposts. Changing prompts or engines for the re-scan makes it impossible to compare with the baseline.
• Changing everything at once. If five changes hit the same prompt group in the same week, you won't know which one worked. Where you can, stagger actions that target the same prompts.
• No change log. Three months later nobody remembers when the pricing page went live.

Cycle 2 and beyond

The retrospective on day 31 produces the next plan. Carry over:

• Actions that moved a metric. Do more of the same, on the next prompt group.
• Actions marked "too early". Keep watching. Don't repeat them yet.
• Findings with no action yet. They're your next candidates.
• New findings from the re-scan. Write them in the Lesson 7 format before you plan.

Keep the baseline method fixed for at least three monthly scans. After that, if you add prompts or engines, version the prompt set (v1.1, v2.0) and report old and new sets separately for one cycle, so the trend line isn't broken.

Our view: after three cycles, most teams know which two or three kinds of action work in their category. That knowledge is worth more than any single month's numbers. Write it down. It's the start of your own playbook.

Exercise

1. List every candidate action for your three findings, then score each on impact, effort and confidence.
2. Pick five, including one access check and every accuracy fix.
3. Write each one as Action, Owner, Due, Metric, Done when.
4. Write your targets before the re-scan, book the weekly check-ins, and share the plan with every owner.

Quiz

1. What should every action trace back to? A finding from your baseline.

2. Why include an access check? Crawlers have to reach your pages before content changes can matter.

3. How many owners should an action have? One.

4. How should you judge the first 30 days? Mainly on actions completed and pages recrawled, with outcomes treated as provisional.

5. Why set targets for prompt groups rather than single prompts? Single prompts often change between runs. In MentionX data, the brand appeared only sometimes in 11 of 28 repeated pairs.

Course complete. Next: Capstone: Break into GEO.

Sources

• Google Search Central, AI features and your website
• OpenAI, Overview of OpenAI crawlers
• Google, Verifying Googlebot and other Google crawlers
• Google Search Central, Sitemaps overview
• MentionX aggregate scan data, 22 August to 29 October 2026 (UTC), 2,154 engine runs, 18 projects, plus a variance cut of 643 runs from 4 September to 19 October 2026. Anonymised and read-only.