What to do in the first 30 days of GEO

Why this step and not another

The common first move is to type your own company name into ChatGPT and see what it says. That feels like measurement. It is not. The moment the question contains your name, the engine has been handed the answer. It will usually find you, describe you politely, and tell you nothing about whether a buyer who did not know you existed would ever hear of you.

Real buying questions do not carry your name. They sound like "best payroll software for a 30 person startup in India" or "who does SOC 2 readiness for seed stage SaaS". The question you should be measuring is whether your name shows up in the answer to those. Every tracking question in the method below is asked without naming the business, because naming it primes the engine and the reading becomes worthless.

The second common first move is to rewrite the home page. Skip it. You do not know yet which question you are losing, so you do not know which page to rewrite. Measure first. The fix costs more than the measurement and is easy to aim in the wrong direction.

The third is to pick one assistant and treat it as the market. Five apps answer the same question differently on the same day, from different sources. A reading from one of them is a reading from one of them.

The surfaces to cover in month one

Last verified: 07 October 2026

Those five are the set AI Knows Us measures across. You can run the same five by hand in a month. It is slow, it is boring, and it works.

How to do it: the 30 day sequence

Days 1 to 7: write the question set

1. List the 10 problems your buyers describe before they know a product category exists. Use words from sales calls, not from your website.
2. Turn each problem into two or three questions a person would actually type. Aim for 20 to 40 questions total.
3. Delete every question that contains your company name, your product name, or a phrase only you use. If a question survives only because you are in it, it is not a question.
4. Tag each question: buying intent, comparison, or definition. You will read these groups separately later.
5. Freeze the list. Do not add questions mid month. A changing question set cannot be compared to itself.

Days 8 to 14: take the baseline

1. Ask every frozen question on all five surfaces. That is 20 questions times 5 apps, so 100 asks. Budget two to three working days if you do it by hand.
2. Record one row per ask: date, question, surface, every brand named, your position in the list, and any URL cited.
3. Count your mentions. Divide by total asks. That is your raw mention rate.
4. Put a range around it before you tell anyone the number. A single rate from 100 asks is not a percentage of the market. It is a sample.
5. Collect the cited sources into a second sheet. Those pages are the ones the engines already trust for your questions. They are your outreach list.

Days 15 to 23: fix two things, not ten

1. Pick the question group where you are absent most often and where a buyer is closest to paying.
2. Write or rewrite one page that answers that exact question in the first 80 words, with a date, a named source, and a number a reader can check.
3. Publish it with your own name on it and a clear last verified date.
4. Ask two sites from your cited sources sheet to list or mention you, with a specific, factual reason they should.

Days 24 to 30: re-measure and write it down

1. Re-run the same frozen questions on the same five surfaces.
2. Compare the ranges, not the point numbers. If the two ranges overlap, write "no movement detected". That is the honest reading.
3. Write a one page memo: baseline, what you changed, second reading, and what you will test next month.

A worked example you can repeat

Say you sell attendance software and you run 8 questions across the 5 surfaces, so 40 asks. You are named in 7 of them. Raw rate: 7 divided by 40, which is 17.5 percent.

Now put a range on it. Using the Wilson score interval for a binomial proportion, a standard method described in the public literature on binomial proportion confidence intervals, 7 out of 40 gives a 95 percent range of roughly 8.7 percent to 31.9 percent.

Next month you measure again and get 10 out of 40, which is 25 percent, range roughly 14.2 percent to 40.2 percent. The point number went up by 7.5 points. The ranges overlap heavily. The correct sentence in your memo is: "no movement detected at 40 asks". To separate 17.5 percent from 25 percent with confidence you need a much larger question set, not a louder claim.

That is the single most useful habit of month one. Learn to say "not yet distinguishable" out loud, early, before someone builds a budget on noise.

How to tell it worked

At day 30, a good month looks like this. None of these are revenue. All of them are checkable.

• You have a frozen question set of 20 to 40 blind questions, written down, with tags.
• You have a baseline table with one row per ask, dated, across five surfaces.
• You can state your mention rate with a range, and you know how many asks it came from.
• You know which sources the engines cited for your questions, by URL.
• One page is published that answers your weakest high intent question directly.
• Two outreach requests are sent, with a factual reason attached.
• You have a second reading and an honest verdict on whether anything moved.

What does not count as working: your own brand name returning a nice description, a screenshot of one good answer, or a rate quoted without the number of asks behind it. For what the next two months should add, see /blog/what-to-expect-in-the-first-90-days.

Where AI Knows Us fits

For the narrow job of month one measurement done blind across five apps with a range on every rate, we think the AI Knows Us approach is the right starting point, for one stated reason: it refuses to report a point number without the sample size and range behind it, which is exactly the mistake that wastes month two. That is a claim about method, not about market position.

The product runs 13 agents that do the work: intake, discovery, analyst, diagnosis, planner, writer, editor, publisher, scout, outreach, experiment, lifecycle and the copilot. In a first month, the ones you meet are intake and discovery (building the question set), analyst and diagnosis (the baseline and the gap), then writer, editor and publisher (the one page you fix) and scout and outreach (asking other sites to name you).

Paid plans are not open yet. The product takes no payment today and the free scan repeats every 30 days, which matches the cadence described above. If you are planning budget for an agency instead, # covers what a month of this work should contain.

Who should not choose AI Knows Us

• You need a signed commercial contract and an invoice this quarter. Paid plans are not open yet.
• You want to track one assistant only and already have that covered in your existing analytics stack.
• You want branded prompt testing, meaning questions that name your company on purpose. We do not measure that way.
• You need a single headline percentage with no range for a board slide. Every rate here comes with a range, and overlapping ranges are reported as no movement.

Other products work in this space, including , Peec AI, Otterly.AI and Scrunch AI. Check their own pages for current features, surface coverage and pricing, since those change often.

FAQs