Start with one user question: a practical GEO experiment

GEO should not start with “we need to write an article.” It should start with a real user question.

This guide is for a team checking how its brand appears in AI answers for the first time. It does not promise that one experiment will change visibility or citations. The goal is a repeatable, comparable starting point.

01 Choose one question

Choose a question that people actually ask and that affects a decision:

• “Which options are worth considering in this field?”
• “How should we choose with a limited budget?”
• “What approach works for a small team?”

Do not cover dozens of keywords on day one. One concrete question makes accuracy easier to judge and change easier to record.

02 Write down the brand facts

Before opening an AI tool, make a fact card: who the brand is, whom it serves, which problem it solves, its limits, and which claims can be verified publicly. Add a primary source such as an official page or product document to each fact where possible.

The card is not a bag of keywords. It is a shared checking standard. If an answer is wrong, the team can see which fact it drifted from.

03 Record the first answer

Ask the same question on two or three AI platforms. Record the date, exact prompt, whether the brand appeared, how it was described, which sources were cited, and any errors or gaps. Keep the full context rather than a screenshot of one flattering sentence.

Use four text states: absent, mentioned but inaccurate, present and relevant, present with verifiable sources. Color can assist scanning, but it must not be the only status signal.

04 Change one variable

Do not rewrite the whole site, publish ten articles, and reposition the brand in the same experiment. Pick one action: add a product fact page, answer one frequent question, or reorganize an existing piece into a clearer answer structure.

Record the URL, date, reason, and expected impact. Without an expectation, the next check cannot explain what happened.

05 Recheck on a fixed schedule

Wait for a fixed interval before checking again. Use the same question, platform scope, and record template. Compare mention status, accuracy, sources, and question coverage, not a single momentary ranking.

No change is still evidence. The content may not have been discovered, the facts may be insufficient, or the question may not fit the brand. Let the gap choose the next action.

06 Experiment record template

User question:
Check date:
Platform:
Brand status: absent / inaccurate / relevant / sourced
Answer and citations:
Fact gap found:
The one change made:
Next check date:

The template is small, but it turns “AI seems to know us” into an observation that can be reviewed. GEO is not about making answers permanently static; it is about keeping the brand clear, relevant, and verifiable as answers change.