How much time does a GEO program need?

With tooling, the owner's involvement is typically a few hours a week plus the monthly review; the content fixes land with whoever owns content anyway. Without tooling, add the run-and-record work from chapter 7 at whatever scale you can sustain honestly.

When should we buy a tool instead of spreadsheets?

Three triggers. When the runs required for trustworthy numbers exceed what a person can do without cutting corners. When leadership starts making decisions on the numbers and the numbers need confidence ranges attached. And when you need to know that your prompt set actually covers your market’s question space rather than hoping it does; measuring coverage means modeling the industry’s questions at volumes only tooling can run, and it is the one part of this discipline a diligent team cannot replicate in-house. Cutting runs to save time is the one economy that defeats the entire purpose.

What does success look like after six months?

Rising share of voice with uncertainty attached on the topics you worked, engines telling your story accurately, a before-and-after list proving cause and effect, and no one in the building treating a single screenshot as evidence anymore.

The GEO Operating Rhythm: Measure, Fix, Re-measure

One audit tells you where you stand. A program changes where you stand. The difference between the two is not effort or headcount; it is rhythm. This chapter is about installing that rhythm so each month’s work builds on the last instead of starting over.

The loop is the program

Everything reduces to one cycle, run on a calendar rather than on enthusiasm:

Measure → pick the biggest gap → fix one thing properly → re-measure → keep what worked.

Monthly is the natural cadence for most teams: fast enough to matter, slow enough for content changes to take effect and for re-measurement to mean something. What kills programs is not choosing a slow cadence; it is having none, so that measurement happens when someone remembers and fixes never get their “after”.

Who owns this

GEO work cuts across content, SEO, product marketing and brand, which in practice means it defaults to nobody. Give it one owner. Not a committee: one person who runs the measurement, keeps the prompt set honest, brings the gaps to whoever owns the relevant pages, and reports the trend.

The owner’s most important power is editorial: they decide what counts as a real change versus noise, and they defend that line when a screenshot lands in a leadership channel. Which brings us to reporting.

Reporting without lying to yourself

Three rules keep a GEO report honest, all of them consequences of Part 1:

Report trends, not moments. Share of voice on your money topics, tracked over months, same method every time. A single month’s reading is weather; the trend is climate.
Report the uncertainty. If the number is 42% give or take 8, say so. The give-or-take is what tells your audience whether this month’s move means anything, and it trains leadership out of reacting to wobble.
Tie fixes to before-and-after pairs. The report’s real product is the growing list of “we changed X, visibility on topic Y moved from A to B, beyond noise”. That list is proof the channel responds to work, and it is what earns the program its budget.

And one standing expectation to set early: model updates reset scores. When an engine swaps its underlying model, answers reshuffle for everyone at once, through no action of yours or your competitors’. A program that has internalized this treats a sudden industry-wide shift as re-baselining, not as crisis or triumph. What survives model updates is what you have learned about which fixes move answers, which is exactly why the before-and-after list matters more than any single score.

Growing the surface deliberately

Start narrow and honest: five to ten topics, two or three engines, your primary market. Expand along whichever axis is answering a real question:

More topics, when the current set is stable and being acted on. New products, new use cases, competitor comparison prompts.
More engines, when your buyers demonstrably use them. Each engine is a separate distribution with its own personality; do not assume your ChatGPT standing transfers.
More personas and markets, when you sell to more than one kind of buyer, or in more than one language. This is routinely where the ugliest gaps hide, because nobody at headquarters ever asks the question the way a buyer in another market does.

The constraint is discipline, not ambition: every expansion multiplies the runs required to keep the numbers trustworthy. Expand when the current scope is measured properly, not instead of measuring properly. At some scale, hand-running prompts stops being a sane use of a human, and the choice becomes automating the loop or narrowing scope. Either is legitimate; drifting into thin coverage of a wide scope is not.

The compounding effect

Run this rhythm for a few quarters and you accumulate assets nobody can shortcut:

a prompt set that actually maps your demand,
a measured history of your category’s answers, including what the engines used to say,
an evidence file of which fixes moved which topics,
and content that got systematically more quotable, which keeps paying into both AI answers and classic search.

That accumulation is the real moat. Any competitor can copy a tactic; they cannot copy your accumulated knowledge of what moves your category’s answers.