Staying Current in AI Search: A Weekly Learning System

#01Why do you need a system rather than a news feed?

AI search changes quickly, and commentary about it changes even faster. Following everything leads to two problems: you spend hours reading, and you react to claims that turn out to be wrong or irrelevant to your market. A system narrows your attention to sources that matter and gives you a way to check claims against your own evidence.

Our view: the best way to stay current is to watch the engines directly. Your own weekly tests will tell you more about your category than most articles about AI search in general.

#02What should you read each week?

Prioritise primary sources. These include official documentation and blogs from search engines and AI companies, such as Google Search Central's guidance on AI features, developer documentation for the assistants your buyers use, and product announcements from those companies. When they change how crawlers work, how sources are cited or how content is chosen, you need to know.

Add a small number of trusted secondary sources: a few practitioners who share evidence and methods, not just opinions, and one or two communities where people compare findings. Unsubscribe from anything that mostly repeats headlines or sells certainty.

Set a time box of about an hour. Skim headlines, read only what affects engines your buyers use or work you're doing, and save the rest for later or never.

#03What should you test each week?

Keep a fixed set of ten to twenty prompts covering your brand, comparisons, recommendations and problem questions in your category. Run them in the assistants your buyers use, ideally on the same day each week, under the same conditions. Record whether you're named, how you're described and which sources appear.

This weekly test isn't your main measurement programme; it's an early warning system. It shows when an engine starts citing different sources, when a competitor appears suddenly, or when an answer about your product goes wrong. Those changes often show up in your own prompts before anyone writes about them.

Rotate in one or two exploratory prompts each week. Try a new type of question, a new feature or a new engine. Note anything surprising.

#04How do you log what you learn?

Keep a simple change log with four columns: date, what changed, source (official, test or secondary) and what it means for your work. Write one or two lines per entry. Over time the log becomes a valuable record: you'll see patterns, such as an engine changing how often it cites sources, and you'll be able to explain shifts in your results.

Tag entries by engine and topic so you can find them later. When something in the log affects a decision, link to it from the decision. This makes your reasoning traceable.

#05How do you separate signal from noise?

Ask three questions of any claim. Who says so, and how do they know? Does it apply to the engines and market I care about? Can I see it in my own tests? Treat claims based on a single screenshot, a small sample or an unnamed source with caution.

When you see a claim that matters, test it. Add a prompt that would reveal the effect if it's real, and run it for a few weeks. Often you'll find the effect is smaller, narrower or absent in your category.

#06How do you share what you learn?

Send a short weekly or fortnightly note to your team: one or two changes that matter, what they mean and any action. Keep it to a few lines. Over time, colleagues will rely on it, and it builds your reputation as the person who knows what's actually happening.

Once a quarter, write a slightly longer summary for leadership: the main changes in the engines your buyers use, what you've adjusted and what you're watching next.

#07How do you fit it into a busy week?

Block two hours in your calendar on the same day each week, and treat it like a meeting. If you're short on time, keep the prompt test and skip the reading; your own evidence is more valuable than commentary. Use a template for the log and the team note so you're not starting from scratch.

#08How do you keep the system from going stale?

Review it every quarter. Drop sources that haven't been useful and add ones that have. Update the prompt set if your products or buyers change, but keep a stable core so trends remain visible. Check whether the engines you test still match the ones your buyers use.

#09From MentionX data: why test several engines weeklyOriginal data

Method. Read-only aggregate queries over MentionX scans from 22 August to 29 October 2026 (UTC): 2,217 engine runs (2,154 completed) across 18 anonymised projects and six engines.

MentionX's anonymised aggregates show how differently engines behave on mentions and citations:

Our view: because engines differ this much, news about one engine tells you little about another. Your weekly test should cover each engine your buyers use.

Limits. A convenience sample of mostly B2B software projects, run through engine APIs, which can behave differently from the consumer apps. Perplexity, Copilot and DeepSeek figures come from a single day (26 October 2026), so treat them as a snapshot, not a trend.

#10Worked example: Ledgerline's Monday hour

Ledgerline is a fictional accounts-payable tool. Figures are illustrative.

The GEO owner at Ledgerline, a fictional accounts-payable tool, blocks two hours every Monday morning. She skims official documentation updates for the three assistants Ledgerline's buyers use, then runs 15 fixed prompts and two exploratory ones.

One Monday, her test shows a forum thread suddenly cited in answers about invoice approvals, repeating an outdated limit. She logs it, posts a short correction in the forum with a link to the docs, and flags it in her team note. Two weeks later, the answers show the correct limit in most runs, in this illustrative example. Her log shows the whole sequence for the next monthly report.

#11What can you do this week?

Choose three primary sources and two trusted secondary sources. Write fifteen prompts. Block two hours in your calendar for next week. Create a change log with four columns and a template for a short team note. For career context, see switching from SEO to a GEO career.

#12What does a weekly routine look like hour by hour?

Here's a simple structure you can adapt. In the first fifteen minutes, skim the official sources on your list for new posts or documentation changes. In the next thirty minutes, read anything that affects engines your buyers use, and note it in your change log. Then spend about an hour running your fixed prompts and the two exploratory ones, recording results in your sheet. In the last fifteen minutes, compare results with last week, add any changes to the log and write your team note.

If something big happens, such as an engine launching a new search feature, swap the exploratory prompts for tests of that feature. Keep the fixed prompts unchanged so your trend stays intact.

#13How do you learn beyond the weekly routine?

Once a month, go deeper on one topic. Read a research paper, work through an engine's developer documentation, or try a small experiment on a page you control. Once a quarter, compare notes with peers outside your company, for example in a professional community, to see whether they're seeing the same changes. Keep confidential details out of these conversations.

Teaching is one of the best ways to learn. Offer a short session for your team on something you've figured out, such as how a particular engine chooses sources. Preparing it will sharpen your own understanding, and questions from colleagues often reveal gaps you hadn't noticed.

#14How do you keep the log useful over time?

Review it every month and highlight entries that turned out to matter. Archive entries that were noise. After six months, write a short summary of the biggest patterns you've seen, such as which engines changed most and which sources became more or less prominent in your category. That summary is useful for planning and for explaining your work to leadership.

Make the log shareable. If you're away, a colleague should be able to read it and understand what's been happening. If you change jobs, a clean log is a strong record of your judgment, although you should leave anything confidential behind with your employer.

#15What should you do when you find a big change?

Confirm it first. Run the affected prompts again over two or three days to check it isn't a one-off. Look for an official announcement or documentation update that explains it. Then judge the impact: does it affect how you're described, which of your pages are cited or how your results should be reported? If it does, tell your manager and the working group promptly, with what you know, what you don't and what you suggest. Note it in your log with the date. Changes caught early are much easier to explain in the next monthly report than changes discovered by leadership first, so speed matters here more than polish. A short, factual message on the same day is usually enough.

#16Common questions

How much time should I spend staying current? About two hours a week: one on reading primary sources and one on testing.

What are the best sources? Official documentation and announcements from the engines your buyers use, plus a few evidence-led practitioners.

How do I know if a claim is true? Check who says it and how they know, then test it with your own prompts.

Should I test every engine? Test the ones your buyers use, and spot-check others quarterly.