How to Implement GEO in 90 Days: The Full Path

Why the order matters

Every stage produces what the next one needs. Skip the baseline and the content plan has nothing to target. Skip the site fixes and the placed material gets discounted when a model comes back to verify. Run the stages out of order and the work still happens — it just cannot be evaluated.

Ninety days is roughly one full loop for a first engagement. It is not a promise about results; it is how long the four stages take when nothing stalls.

Days 1–7: baseline

Write down 15 to 30 questions your customers actually ask, in their wording. Take them from sales and support conversations, not from a keyword tool. Freeze the list, punctuation included.

Run them across every platform, at least three times each, recording four fields per combination: named or not, mention context, other brands present, cited source domains. The full method is in the baseline measurement guide.

Output: one table. Everything afterwards is compared against it.

Days 8–21: site readability

This is the cheapest work in the whole programme and the most commonly skipped. Test whether AI crawlers can actually fetch your pages, whether content survives with JavaScript disabled, whether unknown paths return a real 404, and whether the same facts appear consistently across every page.

Details in the site readability guide. Budget for coordination, not just work — this stage needs an engineer, and engineers have other priorities.

Days 22–70: content and sources, together

The cited-domain column from your baseline tells you where models are taking material from in your category. Produce material that answers the questions you decided to contest, and place it across those source types — not only on your own channels.

Doing content without placement is the most common way a GEO programme produces nothing. See the content and sources guide.

Days 71–90: re-measure

Re-run the frozen question set under the same conditions. Report three numbers per platform — mention rate, share of voice, source coverage — plus what changed and the most likely reason. Then decide what the next loop targets.

Method in the re-measurement and acceptance guide.

Where implementations stall

No owner. The work needs someone internal for a couple of hours a week. Without a named person it stops after the baseline.
Site changes queued behind other work. Raise this before starting, not in week three.
Re-measurement lapses. Rounds one and two happen, round three does not, and after that nobody can say whether any of it worked.

All three are scheduling problems rather than GEO problems, which is exactly why they are worth naming in advance.

FAQ

Questions teams ask before they start

Is 90 days a promise of results?

No. It is how long the four stages take to run once when nothing stalls. What results appear in that time depends on your category, your starting position and how contested the questions are — which is exactly what the baseline is for.

Can the stages be run in parallel?

Content and placement run together, and the site fixes can overlap with them once the specification exists. The baseline cannot be parallelised with anything, because everything else is compared against it.

What if we only have budget for part of it?

Run the baseline and the site fixes. Those two are the cheapest and they are prerequisites for everything else; without them, content spend is not measurable and placed material gets discounted on verification.

Who should own this internally?

One named person with a couple of hours a week, positioned close enough to marketing to confirm claims and close enough to engineering to get site changes scheduled. Without a named owner, engagements stop after the baseline.

What happens after the first 90 days?

The same loop, targeted by what the first round showed. The question set stays frozen; what changes is which questions are contested next and which source types the placement work concentrates on.

Is 90 days the minimum?

It is the shortest span in which one full loop including re-measurement fits. A shorter engagement can still produce work, but it cannot produce a comparison, which means it cannot be verified.

The 90-day GEO implementation path