How to Test GEO in 90 Days: Measuring AI Citations

The short answer

A 90-day GEO test has three 30-day phases. During that time, you measure whether AI models such as ChatGPT, Perplexity, Claude and Google AI Overviews use your content and brand as a source. In month one, establish a baseline and build the foundations with well-structured content. In month two, publish consistently and develop topical clusters. In the final month, assess which citations are increasing and refine your approach based on the results. Without a defined process, you are simply publishing content and hoping it pays off.

Key takeaways

• A GEO test means very little without a baseline. In week 1, record how often your brand appears in ChatGPT, Perplexity and Google AI Overviews using the same targeted prompts each time.
• Allow at least 90 days. Crawlers and AI models need time to index content and recognize repeated mentions as a trustworthy signal.
• A topical cluster performs better than disconnected articles. Build one in-depth pillar page supported by 6 to 10 related articles that link to one another naturally.
• Structured data, including FAQ and organization schema, plus a direct answer near the start of each article, can improve your chances of earning a useful citation.
• Measure every two weeks using the exact same prompts. Daily checks mostly create noise because AI responses and source indexes do not refresh continuously.

Why GEO tests often stall after just two weeks

Many marketing managers start with the best intentions. They publish ten articles, ask ChatGPT a week later whether their brand is mentioned, see no result and conclude that GEO does not work. In reality, they often stop just before the test has had a fair chance to deliver anything meaningful.

The issue is usually not content quality. It is unrealistic expectations and inconsistent measurement. AI models rely on sources they consider reliable and consistently relevant. A newly published article is not automatically picked up overnight. Crawlers must first discover and process the page. ChatGPT and Perplexity also rely in part on their own source indexes, which are not updated constantly. Search Engine Journal notes that citations in generative search engines often come from repeated mentions across multiple pages and domains, rather than from one exceptional article.

Many small and mid-sized teams also lack a consistent measurement process. One week, they use a different question. The following week, they test a different model. That does not create a like-for-like comparison. A 90-day GEO test solves this by using fixed phases, scheduled measurement points, consistent prompts and clear KPIs.

Why AI visibility requires a 90-day window

GEO results do not appear overnight. They build gradually through consistent publishing and repeated indexing. With SEO, you may see ranking movement within a few weeks. AI search adds another layer: it must not only find your page, but also consider it credible enough to cite ahead of other sources.

That is why 90 days is a realistic minimum:

• Models process sources at different speeds. Perplexity and Google AI Overviews may pick up new pages relatively quickly. Models that rely partly on periodically updated training data often take longer.
• Brand recognition grows through repetition. An AI model needs to encounter your brand repeatedly in connection with the same topic before it begins to treat you as an expert source.
• A cluster needs enough depth. One article is rarely enough. Eight to ten topically connected articles around a central theme send a much stronger signal.

This aligns with HubSpot's research on AI search: brands that publish consistently on a subject are cited more often than those that post only occasionally. Want to compare different approaches first? Read which comparison actually helps you choose the right SEO tool.

How to test what AI models cite in 90 days

The foundation of a strong GEO test is straightforward: follow a consistent sequence over three months and do not skip measurement points.

Step 1: Create a baseline using ten fixed prompts

Write down ten questions a prospective customer would realistically ask ChatGPT, Perplexity or Claude about your product category. For every question, record which brands appear, whether your brand is included and which sources the response draws on. This is your starting point. Without a baseline, you cannot demonstrate whether anything has changed after 90 days.

Step 2: Choose one core topic and build a cluster around it

Choose the topic you want to be known for. Then create one comprehensive pillar page and support it with six to ten in-depth articles. Link the articles together in a logical way. Standalone articles on unrelated subjects do little to reinforce one another. A platform such as Launchmind's GEO solution can help you build these clusters methodically, rather than producing articles in isolation.

Step 3: Write articles that answer the question immediately

Start every article with a clear 80 to 120-word answer to the question in the title. AI models favor passages that make sense even when separated from the rest of the page. Add an FAQ section and organization schema where appropriate, so crawlers can better interpret the content and its structure.

Step 4: Publish consistently, not in one big burst

Publishing twenty articles in one week and then going silent for three months rarely works well. A pace of two to three articles per week throughout the full period is far more effective. It builds your topical authority step by step. In practice, many teams find this difficult alongside their day-to-day work. That is why some small and mid-sized businesses choose an AI colleague that publishes content to a consistent schedule.

Step 5: Repeat the same measurement every two weeks

Use the same ten prompts from your baseline and record every outcome. Look beyond whether your brand is mentioned. In what context does it appear? Is your brand presented as a specialist, one of several options, or not at all? A simple spreadsheet with the date, prompt, model and result is enough. The discipline of maintaining it matters more than an elaborate measurement tool.

Step 6: Identify which articles are generating citations

By around day 60, the first patterns often emerge. Which articles receive strong impressions and clicks in Google Search Console? Which topics recur in AI model responses? Use that information to refine your strategy. Consolidate or rewrite weaker articles. Develop the topics that are clearly gaining traction. This is also where tools differ: some rely only on estimated search volume, while an approach based on real Search Console data shows more quickly what is working.

Step 7: Show progress across the full period

After 90 days, do not present only a final figure. Compare your baseline, the mid-point measurement around day 45 and the final measurement side by side. For a CMO, a clearly upward trend is often more persuasive than a single end result. Want to place this within a complete quarterly plan? See what should be included in a 90-day GEO roadmap for a small or mid-sized team.

Get started:

• Create ten fixed prompts and test them in ChatGPT, Perplexity and Claude this week.
• Choose one core topic for which you want to build visibility within 90 days.
• Schedule two to three publications per week.
• Put a recurring measurement session in your calendar every two weeks, separate from other marketing work.
• Decide in advance who will prepare the day 90 report and who will receive it.

How to avoid unreliable results

Many GEO tests fail because the measurement setup is messy, not because the content is poor. Follow these three rules to keep your results useful.

Use the same prompts in the same order every time. AI models are sensitive to wording. The questions "Who is the best provider of X?" and "Which provider of X is the most reliable?" can produce very different responses, even though they appear to ask roughly the same thing. Save the exact wording in a document everyone uses.

Test in a fresh session or an incognito window. Models that retain conversational context may factor in previous questions, which can skew your results.

Compare your brand with competitors as well. If you are mentioned after 90 days but always appear as the third or fourth option, that is different from being presented as the top recommendation. Record both the position and context of every mention.

Mistakes that can derail a GEO test

The biggest mistake is stopping when early results do not appear. Between day 20 and day 30, most teams see very little difference. That is often when enthusiasm drops and the test is abandoned, even though the impact of a topical cluster frequently becomes more visible only after day 45.

A second mistake is treating SEO rankings and AI citations as the same metric. A page can rank highly in Google without ever being mentioned by ChatGPT. AI models also look for different signals, including direct answers, clear structure and strong brand recognition as a source. Want to assess the two separately? Read the review of Ahrefs Brand Radar for GEO.

A third mistake is ignoring older content. Teams often focus all their energy on new articles while existing pages contain outdated, unclear or conflicting information that can undermine trust. Cleaning up, consolidating or removing old content should be part of the same strategy.

Finally, using several measurement tools at once without one governing process does not help. If you use three platforms side by side, you can quickly end up with data that is not comparable. Choose one measurement method for the entire test period and do not switch halfway through.

Get started:

• Schedule a review on day 30 before deciding whether to stop.
• Check whether older articles contradict each other on the same subject.
• Document which tool you use for SEO and which method you use to track AI citations. Keep those reports separate.
• Choose one measurement method for the full 90-day period and document that decision.

Frequently asked questions

How many articles do you need for a reliable 90-day GEO test?

Aim for at least eight to ten articles around one central topic. This gives AI models enough connected information to associate your brand with that subject. With fewer than six articles, the topical foundation is often too thin to draw clear conclusions within 90 days.

Which tools can you use to track AI citations without doing everything manually?

Specialist platforms are available for tracking AI visibility. Many small and mid-sized teams combine this type of measurement with a content solution that publishes and adjusts based on Search Console data, such as Launchmind. That way, content production and measurement do not become two disconnected workstreams. Want to compare the options first? Read GEO optimization tools for small and mid-sized businesses in 2026.

What does a 90-day GEO program typically cost?

Costs vary significantly depending on the approach. A freelance writer will often charge per article. An agency typically works on a fixed monthly retainer that can run into thousands of euros. An automated platform usually charges a subscription fee or a price per publication. Always ask what is included. Writing, publishing, measurement and optimization are four separate services, and not every provider covers all of them.

Why does ChatGPT mention my competitor but not my brand when our content is similar?

It often comes down to structure and repetition, not content quality alone. Competitors that are consistently associated with the same subject in several places, such as their own website, industry publications and reviews, build a stronger brand profile than companies that only write about themselves on their own blog.

Can you run the test in 30 or 60 days instead?

You can measure early signals within that time frame, but drawing a reliable final conclusion is risky. AI models need time to process repeated mentions and new sources. Use a day 45 measurement to refine your approach, not as a definitive verdict on GEO.

Conclusion

A 90-day GEO test is not an excuse to ask for patience. It is a realistic time frame in which AI models can begin to recognize your brand as a trustworthy source. If you quit after two weeks, you have not really run the test. A solid baseline, a connected content cluster, a consistent publishing rhythm and biweekly measurement are what separate informed decisions from guesswork.

For small and mid-sized teams, the challenge is often not strategy but execution. Who writes two to three articles every week? Who maintains the topical connections? Who measures consistently? Launchmind was built to keep that process moving. The platform writes, reviews and publishes content to your own website every day, for Google and AI search engines such as ChatGPT and Perplexity. It also adapts based on real Search Console data, not gut instinct. Want to explore what a 90-day program could look like for your business? Book a no-obligation consultation or see how to get your first articles live.

About the company

Launchmind is the AI colleague that writes, reviews and publishes SEO content to your own blog every day, in eight languages. The platform continuously improves based on real Search Console data. Launchmind is designed for marketing managers, entrepreneurs and CMOs at small, mid-sized and scale-up businesses that know content works but struggle to execute it consistently themselves.