AI Search Visibility
Weekly AI Search Review: A 30-Minute Team Agenda
A timed 30-minute agenda for your weekly AI search review, with the Noma view behind each item, a noise-band test and an action triage table.

A weekly AI search review is a 30-minute meeting that checks whether AI engines still name your brand, which pages they cite, how competitors moved, what new gaps appeared and who owns each fix. Weekly is the right default because single AI answers are noisy and monthly checks arrive too late. Noma by Nurdd supplies the data for each agenda item across 13 AI engines.
A weekly AI search review is a 30-minute team meeting that checks whether AI engines still name your brand, which pages they cite, how competitors moved, which new prompt gaps appeared and who owns each fix. Weekly is the right rhythm because daily checks mostly capture random variation, while monthly checks arrive after the damage is done.
Most teams that track AI answers already have a dashboard. What they lack is a fixed meeting that turns the dashboard into decisions. This guide gives you the agenda, the minute-by-minute timing, the Noma view that feeds each item, a noise-band test you can run on your own history, and a triage table for the actions that come out of the room.
Key takeaways
- Review AI search visibility weekly for your core buyer prompts, and move only a short watchlist to daily during launches or PR issues.
- Use five timed agenda items: noise check, citations, competitor movement, new gaps, and owned actions. Anything that does not feed one of them stays out of the room.
- Agree a noise band before the meeting. One answer is a sample, not a trend, so only act on changes that repeat, span engines or hit a high-value prompt.
- Finish with no more than three actions, each with a named owner and a Friday checkpoint.
- Prepare the numbers 10 minutes before the meeting, not during it. Use the GEO readiness scorecard to check your starting point if you have no baseline yet.
How often should a marketing team review AI search visibility?
Weekly is the default for the core set of buyer prompts. MaxAEO's AI search monitoring cadence guide reaches the same conclusion for most B2B SaaS teams, reserving daily checks for launches, reputation risk and fast-moving competitor campaigns, and monthly checks for stable, low-risk prompts.
The table below is our own synthesis of that guidance, reduced to the decision a team actually makes: which rhythm for which job.
| Rhythm | Use it for | Who attends | Risk if it is your only rhythm |
|---|---|---|---|
| Daily (watchlist only) | Product launches, PR issues, inaccurate brand descriptions, a competitor campaign on your category prompts | One owner, async | Noise without action if nobody can fix things daily |
| Weekly (30 minutes) | Core category, comparison, branded and buyer-intent prompts | SEO or growth lead, content owner, PR or comms owner, a decision-maker | Can miss a sudden drop unless an escalation rule exists |
| Monthly | Trend line for leadership, low-risk educational prompts | Marketing leadership | Too slow to catch a competitor displacing you |
| Quarterly | Prompt library cleanup, engine coverage, competitor set, budget | Channel owner and leadership | Not a detection tool |
If you want a deeper treatment of the daily end of this table, see our guide to AI search monitoring alerts, and for how many prompts to put on each rhythm, AI search prompt tracking: how many prompts you need.
Why is a weekly AI search review hard to run well?
The hard part is not the meeting. It is that the data underneath it moves for reasons unrelated to anything your team did. Two published studies show how large that background movement is.
SparkToro and Gumshoe had 600 volunteers run 12 brand-recommendation prompts through ChatGPT, Claude and Google's AI surfaces 2,961 times, as summarised by ThriveStack. SparkToro's own write-up states there is a less than 1 in 100 chance that ChatGPT or Google's AI, asked 100 times, gives the same list of brands in any two responses (SparkToro).
Profound compared about 80,000 prompts per platform between 11-13 June and 11-13 July 2025 and measured citation drift, the share of cited domains in July that were absent in June for the same prompt. It found 59.3% for Google AI Overviews, 54.1% for ChatGPT, 53.4% for Microsoft Copilot and 40.5% for Perplexity (Profound, July 2025 comparison).
Put together, these findings define the job of the weekly review. It must separate repeated, cross-engine movement from random variation, and it must do so quickly enough that the team can still act. That is why the agenda below starts with a noise check and not with a tour of charts.
What should a weekly AI search review meeting cover?
The meeting should cover five things in a fixed order: whether movement is real, which citations changed, how competitors moved, which new gaps opened and what the team will do about them. Parse's weekly AI visibility review uses a similar four-block, 30-minute shape and says openly that 30 minutes is a suggested format, not a measured optimum. Treat the timings below the same way: adjust them after four weeks of your own use.
The agenda adds one thing the published templates leave out: for each item, the Noma view the lead should have open or pre-pulled. Noma by Nurdd tracks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, polled daily, so every item draws on a week of runs rather than one lucky or unlucky answer.
| Minutes | Agenda item | Question answered | Noma view that feeds it | Output |
|---|---|---|---|---|
| 0-5 | 1. Noise check | Did the share of tracked prompts that name us move outside our noise band? | The record of whether the brand is named, per engine | One line: signal, watch or noise |
| 5-12 | 2. Citations | Which pages are cited when we are named, and which sources appeared or vanished? | Citation sources, showing the page each citation comes from | List of gained and lost citation sources |
| 12-18 | 3. Competitor movement | Who gained or lost ground on the prompts we care about? | Competitor comparison | One sentence per competitor that moved |
| 18-24 | 4. New gaps | Where are we absent that we were not absent last week? | The named or not-named record, plus citation frequency by category | Ranked gap list, top three only |
| 24-30 | 5. Actions | What do we do this week and who owns it? | Noma recommendations to improve AI visibility | Up to three tasks, each with owner and Friday checkpoint |
Item 1 (minutes 0-5): Is the movement real?
Open with one simple figure that the lead has worked out beforehand: the share of tracked prompts where Noma recorded the brand as named, compared with previous weeks. Set it against your noise band, which you establish once using the test in the next section. Say out loud one of three verdicts: noise, watch or signal. If it is noise, the meeting does not stop; it moves to the items where specific pages and competitors can still show something actionable.
The reason this item comes first is discipline. Without it, the group spends the first ten minutes debating whether a three-point wobble means anything, which is the conversation the SparkToro and Profound numbers say you cannot settle from one answer.
Item 2 (minutes 5-12): Which citations changed?
Citations are where most explanations live, because an engine that stops naming you has usually stopped reading a page or a third-party source that used to support you. In Noma, the citation view records which page each citation comes from, so you can list sources gained and lost since last week.
For each lost source, ask three questions: did we change or remove that page, did the third-party site change it, or did a competitor replace it? If you have never audited which of your own pages get cited, our guide to the Google AI Overviews citations audit gives a seven-step method that carries over to other engines.
Item 3 (minutes 12-18): Which competitors moved?
Use Noma's competitor comparison to see which rivals gained or lost presence on your tracked prompts. Limit this to the competitors that moved, and write one sentence for each: what changed, on which engine, and the most likely cause.
Do not turn this into a full competitive review. The goal is to catch a displacement early, such as a rival now named first on a comparison prompt that matters to revenue. The measurement logic behind these comparisons is covered in AI search share of voice: how to measure it.
Item 4 (minutes 18-24): Which new gaps opened?
A new gap is a prompt where your brand was named last week and is not named this week, or a prompt that has never named you but now matters because buyers ask it. Noma's record of whether the brand is named shows where you are absent, and citation frequency by category helps you see which topic areas are thin.
Rank the gaps by business value and keep only the top three. If you are unsure which prompts deserve a place on the list at all, start with which prompts to track for AI search visibility.
Item 5 (minutes 24-30): What are we doing about it?
Convert the three gaps into at most three actions. Each action needs a verb, a page or source, a named owner and a Friday checkpoint. Noma's recommendations give the lead a starting list, and the team decides which ones are worth this week's capacity.
Parse's template limits the team to one bet per channel for the same reason: a meeting that produces seven tasks tends to produce none. The triage table later in this guide helps you choose which three.
What should the lead prepare before the meeting?
The lead should prepare the data in 10 minutes the morning of the meeting, so nobody scrolls through a live dashboard while the clock runs. Follow this checklist in order.
- Open Noma and work out the share of tracked prompts where the brand is named this week, then put it beside the previous four weeks in your meeting document.
- Write the verdict for item 1 (noise, watch or signal) against your noise band before anyone joins.
- List the citation sources that appeared or disappeared, with the page each one points to.
- Note which competitors moved in Noma's competitor comparison, on which engine and on which prompt group.
- Pick the three most valuable prompts where the brand is newly not named, and write one hypothesis for each: content, citation or competitor.
- Pull the matching Noma recommendations and shortlist no more than five candidate actions.
- Check last week's three actions: done, not done, or blocked. Carry only the blocked ones into the room.
- Paste everything into one running document, with a five-line summary at the top. Do not prepare slides.
If you have never scored your site's readiness for AI answers, run the free GEO readiness scorecard once so the first review has a baseline to refer back to.
How do you set a noise band for your own data?
You set a noise band from your own history, not from a number in someone else's article. Parse suggests presuming a small move to be noise on a 30 to 80 prompt set, and says thresholds should come from your own engine and prompt history. That is a sensible starting point, but it is a rule of thumb from one vendor, so test it against your own weeks.
Here is a test you can run in the first month without any special tooling.
- Record the share of tracked prompts that name your brand, taken from Noma, on the same weekday for four consecutive weeks, changing nothing on your site in that period if you can.
- Take the highest and lowest of the four readings. The gap between them is your provisional noise band.
- Repeat the check by engine for the two engines your buyers use most, because Profound's drift figures differ by platform, from 40.5% for Perplexity to 59.3% for Google AI Overviews.
- Write the band into the top of the meeting document and leave it alone for a quarter. Revisit it when you add many prompts or change the engine list.
The principle behind the test is the one the SparkToro research points to: report the share of runs that name you, not a single answer, and use many runs per prompt. Because Noma polls daily across its engines, a week of data already gives you several samples per prompt, which is far better than one answer but still a modest sample, so keep your noise band wide at first.
Which verdict do you give a change?
MaxAEO's guidance says a change becomes meaningful when it repeats over time, appears across multiple engines, affects a high-value prompt, or comes with a citation or source change. The decision table below turns those four tests into a verdict.
| What you see | Repeats for 2+ weeks? | Across 2+ engines? | High-value prompt or new citation source? | Verdict |
|---|---|---|---|---|
| Small move inside the noise band | No | No | No | Noise: log it, no action |
| Move outside the band on one engine | No | No | No | Watch: re-check next week |
| Move outside the band, one engine | Yes | No | Either | Signal: assign a bet |
| Any move | Either | Yes | Yes | Signal: assign a bet this week |
| Brand disappears from a top revenue prompt, or an engine states something false about you | Not required | Not required | Yes | Escalate: same-day standup, not the weekly meeting |
Parse also names escalation triggers: a large single-platform drop week over week, a source that contradicts a fact about your brand, or a competitor displacing you on a top-five revenue prompt. Those thresholds are Parse's rules of thumb, so set your own, but keep the idea that some events skip the weekly queue.
How much data does a weekly AI search review actually cover?
A worked example shows why a human cannot do this by hand. MaxAEO's cost formula is prompts x engines x runs x monitoring days = answer checks, and its example of a weekly operating model, 60 prompts across 5 engines, comes to about 1,200 checks per month.
Now apply the same arithmetic to Noma's published coverage. Take 60 tracked prompts across 13 AI engines, polled daily. That is 60 x 13 = 780 prompt-engine pairs per day, and 780 x 7 = 5,460 answer samples in the week the meeting reviews. The 60 is our illustrative prompt count, and the 13 engines and daily polling come from Noma's published description.
Doing 5,460 checks by hand each week (780 a day) is not realistic, and 60 prompts is a modest set. This is the practical case for software behind the agenda: the meeting reviews summaries of thousands of samples, and the people spend their 30 minutes on judgement. Our comparison of manual prompt checks versus software covers where a manual log still makes sense, for example in the first week while you learn what your answers look like.
Which action should the team pick for each type of gap?
Match the gap type to the action, then pick the three with the highest business value. This table is our own triage framework; use it in item 5 so the team does not debate from scratch every week.
| Gap you found | What it usually means | Typical weekly action | Owner | Guide |
|---|---|---|---|---|
| Engines cite competitors, review sites or forums, not you | Source gap | Pitch or update one third-party source the engines already cite | PR or comms | Why your competitor is cited and you are not |
| Engines describe the brand wrongly or mix it up with another | Entity gap | Fix one brand profile, description or listing | Brand or web | Entity consistency checklist |
| Your page ranks on Google but is not cited | Format or evidence gap | Restructure one page with a direct answer, a table and a sourced figure | Content | Ranking on Google but not cited by AI |
| One engine drops you, others do not | Engine-specific behaviour | Check crawler access and that engine's sources before changing content | SEO or web | AI crawler user agents list |
| Absent from a prompt nobody has ever answered on your site | Coverage gap | Publish or extend one page that answers that question directly | Content | AI visibility action plan |
One rule keeps the table honest: change one thing per prompt group per week, so next week's data can tell you whether the change worked. If you want to test changes more formally, GEO experiment design explains how to set up a before-and-after test.
Why does Noma belong in this meeting?
Noma by Nurdd is an AI search visibility dashboard that tracks whether a brand appears in answers from 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, polled daily. It records whether the brand is named and which page each citation comes from, shows citation frequency by category and competitor comparison, and gives recommendations for improving visibility. That is the data each of the five agenda items above needs.
Book a demo at https://www.nurdd.club/demo to see how Noma tracks your brand across 13 AI engines.
Noma has a 3-day free trial. A card is taken at signup and billing starts on day 3, so you are not charged if you cancel before then. Three days covers a first look at your data, and our AI visibility tool trial plan suggests what to test in that window. You can read more on the product page at nomahq.app.
How would three different roles run the weekly review?
The agenda stays the same, but the person leading it and the question they care about most changes. These scenarios show how to adapt it. They describe how a role would use the agenda, not customer results.
The in-house SEO lead who already reports on Google rankings
This lead owns the meeting and brings the pre-processed document. Item 2 matters most, because the lead can compare which pages Noma shows being cited with the pages that rank in Google, and flag pages that rank but are not cited. The weekly output usually leans toward content restructuring, with one action for the content owner and one for PR.
This lead should also keep the AI numbers beside the organic ones. Our guide to AI search metrics versus SEO metrics shows what to report together.
The agency strategist running several client brands
An agency strategist runs one short review per client, or one combined session with a per-client block of five minutes. The strict noise band and the same five items keep reviews comparable across clients. The output that matters most is item 5, because the actions turn into the line items the strategist reports back each month.
For the monthly layer, see AI search reporting for agencies: what to send monthly, and for tools suited to many brands, tracking multiple brands in AI search.
The founder or lone marketer with no team
A single person can still run the review, in 15 to 20 minutes, by merging items 3 and 4 and cutting actions to one. The discipline of a fixed slot matters more than the length. Put the slot in the calendar, write the noise verdict first, then pick one action.
If you have no capacity for the actions themselves, the Nurdd team also runs answer engine optimization as a managed service, measured in Noma. The related generative engine optimization services shape the content and sources the engines read.

What mistakes kill a weekly AI search review?
Weekly reviews often fade, and the causes are predictable. Design them out in the agenda before the first meeting.
- Reacting to every wobble. Without a noise band the team chases variance, which the SparkToro and Profound findings make unavoidable at the single-answer level.
- Changing the prompt list every week. The trend line breaks if the questions keep moving. Edit the list quarterly, as MaxAEO recommends for governance.
- Collapsing all engines into one number. Profound's drift ranges from 40.5% to 59.3% by platform, so a stable total can hide a move on one engine. Read the engines separately in item 1.
- Tasks without owners. If an action has no name and no Friday date, it does not exist.
- Showing raw dashboards in the room. The lead should pre-process data into the five items, as in the checklist above.
- Ignoring personalised variation. Answers can differ by user and context, so compare like with like. See whether personalization changes AI answers for how to account for it.
- Skipping the weekly record. A one-paragraph log per week turns twelve meetings into a quarterly story. The monthly report template shows how to roll weeks up.
How do you roll the weekly review into monthly and quarterly reporting?
Treat the weekly meeting as the operating layer and write a five-line entry each time: verdict, citations changed, competitor moves, gaps, actions. At month end the lead merges four entries into the executive update. At quarter end the team revisits the prompt list, competitor set and engine coverage.
That structure also fits a wider plan. If you are building one from scratch, the AI search marketing plan: 90-day template places the weekly review inside a quarter, and what executives need to see in an AI search visibility dashboard covers the monthly view. For how Noma fits with other Nurdd work, see how Nurdd works.
How we researched this
We reviewed published guidance on AI search monitoring cadence and weekly review agendas from vendors and independent publishers, plus published studies on answer inconsistency and citation drift. Sources were checked in November 2026; where we could not confirm a publication date, we name the source without one. We wrote the agenda, triage table and noise-band test ourselves. We did not run hands-on tests of rival tools and did not interview customers; the Noma descriptions come from Nurdd's own product facts.
Frequently asked questions
How often should a marketing team review AI search visibility?
Weekly is the best default for the core set of buyer prompts. Add a short daily watchlist only for launches, PR issues or a competitor campaign, and keep low-risk educational prompts on a monthly rhythm. Review the prompt list, engines and competitor set once a quarter.
What should a weekly AI search review meeting cover?
It should cover five items in a fixed order: whether the movement is outside your noise band, which citations changed, how competitors moved, which new gaps opened, and the up to three actions with owners. Anything that does not feed one of those items should stay out of the meeting.
Why not review AI search visibility daily?
Daily numbers mostly reflect random variation. SparkToro found a less than 1 in 100 chance that two responses to the same prompt give the same brand list, so a single day's change is hard to read. Daily review makes sense only for a short watchlist where a late discovery would cost you.
How long should the meeting be, and who should attend?
Thirty minutes is a suggested format, not a proven optimum, so adjust it after a month. Keep the room small: the lead who owns the document, the content owner, the PR or comms owner and a decision-maker who can clear blockers.
How many actions should come out of the meeting?
No more than three, each with a named owner and a Friday checkpoint. Changing one thing per prompt group also lets you see in next week's data whether the change had an effect.
What does Noma show for a weekly review?
Noma tracks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, polled daily. It records whether your brand is named, which page each citation comes from, citation frequency by category and competitor comparison, plus recommendations. Those cover each agenda item in this guide.
Is Noma free to try?
Noma has a 3-day free trial. A card is taken at signup and billing starts on day 3, so you are not charged if you cancel before then. Prices are in USD.
How can I see this for my own brand?
Book a demo at https://www.nurdd.club/demo to see how Noma tracks your brand across 13 AI engines, so your first weekly review starts with real data.