# When is a change in AI visibility real? Look past 4 points a week

Magnus Windeløv, Head of Marketing · Nov 4, 2026

Original research from 90 days of daily visibility for 151 brands: how much the number moves on its own, what a real change looks like, and a simple rule for when to react.

Nine in ten week-to-week moves in AI visibility are smaller than 4.3 points. We followed the daily visibility of 151 brands for 90 days. A move of 1–3 points in a week is normal and tells you very little. Single-day jumps of 10 points happen to most brands at least once a quarter. A move of 4–5 points or more that holds for two weeks is the one worth explaining.

I spent years reporting keyword rankings to clients. Rankings wobble, and every report came with the same question: is this drop real? AI visibility wobbles more, and the questions are the same: a Head of Marketing sees the number fall 6 points on a Tuesday and has to decide whether to raise it with the CEO or wait.

Our [State of AI visibility report](https://www.get3rd.com/blog/state-of-ai-visibility-2026) showed that the typical brand's daily score swings 12.5 points between its best and worst day in a month. This article turns that into a rule you can use: what's normal week to week, and when a change is real.

## The short answer

- **A normal week moves about 1 point.** The median week-to-week move was 1.1 points.
- **Nine in ten weekly moves are under 4.3 points.** Treat anything smaller as noise unless it keeps going.
- **Single days are noisy.** The typical brand's biggest one-day jump in a quarter was 10 points.
- **Fewer answers, more noise.** Brands with about 60 answers a day moved twice as much day to day as brands with about 290.
- **So report weekly or monthly,** show the normal band next to the number, and only investigate moves that hold.

## What we found

We tracked daily visibility for 151 brands with at least 35 days of 30 or more answers, from 5 August to 2 November 2026. That's 1,276 week-to-week changes.

| Weekly move | Share of all weekly moves |
| --- | --- |
| Under 1.1 points | 50% |
| Under 2.8 points | 80% |
| Under 4.3 points | 90% |
| Under 6.4 points | 95% |

- **Every brand has one big week.** The largest weekly swing in 90 days was 3.7 points for the typical brand. For one brand in ten, it was 9 points or more.
- **Single days swing much more than weeks.** The typical day-to-day move was 2.1 points. Each brand's largest one-day jump in the quarter was 10 points at the median, and 25 points for one brand in ten.
- **Sample size matters.** The more answers a brand gets per day, the steadier its number:

| Answers per day (median) | Typical daily move | Typical weekly move | Biggest week in 90 days |
| --- | --- | --- | --- |
| About 60 | 3.2 points | 1.7 points | 4.0 points |
| About 140 | 2.1 points | 0.9 points | 3.4 points |
| About 290 | 1.6 points | 1.0 points | 3.2 points |

## Why visibility moves when nobody touched anything

AI assistants don't give the same answer twice. Three things move the number without any change on your side.

1. **Answers vary.** Ask the same question twice and an assistant can name different brands. Every daily score is a sample, and samples move.
2. **Models change.** When an assistant switches to a new model, its answers change overnight. A jump on that day measures the new model, not your work.
3. **Sources change.** A review site updates its list, or a news article drops out of the index. Your visibility moves because the assistant's reading list moved.

None of this means the number is wrong. It means one day, or one week, isn't a trend. The 3RD docs put it plainly: trend direction beats any single day's number.

## A simple rule for when to react

| What you see | What it most likely is | What to do |
| --- | --- | --- |
| A move of 1–3 points in a week | Normal variation | Nothing. Keep watching the trend. |
| One day jumps or drops 10 points | A one-day swing, or an assistant changing models | Check whether an assistant changed models that day. Wait for the week. |
| A move of 4–5 points or more that holds for 2+ weeks | A real change | Find the cause: which assistant, which topic, which sources. |
| A slow drift in one direction over a month | A real trend | Treat it as a signal, even if no single week looked big. |

If your brand gets fewer than about 100 answers a day, widen the bands. Small samples swing more.

## What to do about it

Today, most teams handle the number in one of three ways:

- **React to the daily number.** Every dip becomes a Slack thread. Most of them are noise.
- **Check once a month by hand.** A single screenshot can't tell a real drop from a bad day.
- **Ignore it.** The number moves so much that nobody trusts it, so nobody acts on it.

## Step 1: Report weekly or monthly averages

A daily number invites a conversation about noise. Report the weekly or monthly average, with the trend line. If someone asks about one day, show them the week.

**In 3RD:** the visibility chart on the Overview shows the trend over a week, a month or longer. Switch the period to a month or more to see how visibility has actually developed.

![3RD Overview: daily visibility over 90 days, with a marker on the day an AI assistant changed models. Demo workspace; brand names blurred.](https://www.get3rd.com/media/3rd-data-overview-trend-90d.jpg)

*Daily visibility over a quarter, with the trend. Demo workspace; brand names blurred.*

## Step 2: Show the band next to the number

"We're at 24%, and a normal week moves us about a point" frames the number better than the number alone. Use the table above, or work out your own: look at your last 12 weeks and note how far the number usually moves.

## Step 3: Check for model changes before you investigate

When the number jumps or drops on one day, check first whether an assistant changed models that day. If it did, the jump measures the new model. Look at the following week before you draw conclusions.

**In 3RD:** days when an assistant switched to a new model are marked on the visibility chart, so you can see whether a jump came from the assistant or from you.

## Step 4: When a move holds, find the cause

A move of 4–5 points or more that holds for two weeks deserves an explanation. Break it down: which assistant moved, which topic, which market. Then open the answers and the sources behind them.

**In 3RD:** every metric can be filtered by topic, market, audience and model. Diagnostics links the metric that moved to its likely root cause and the recommended fix, with the evidence behind it.

## Step 5: Judge your own work over weeks, not days

AI assistants re-read the web on their own schedule. Expect days to weeks between shipping a fix and seeing it in answers. Compare the questions you worked on with the ones you didn't, over a month. Our guide on [how to prove a GEO change moved your AI visibility](https://www.get3rd.com/blog/prove-geo-change-moved-ai-visibility) covers this in detail.

**In 3RD:** a Playbook tracks the trend for the prompts a piece of work was meant to move, next to the tasks you shipped.

## What 3RD doesn't do

- It doesn't remove the noise. AI answers vary, and an honest number shows that.
- It doesn't claim a task caused a rise. It shows the trend next to the work you shipped.
- It doesn't decide what's worth escalating. The bands in this article help you make that call.

## Common mistakes

- **Escalating a one-day drop.** Wait for the week.
- **Celebrating a one-day jump.** Same rule.
- **Comparing two screenshots.** Two single answers say nothing about a trend.
- **Missing a slow drift.** A month of small moves in one direction is a real change, even if no single week looked big.
- **Judging work after a week.** Give it a month, and compare it with the questions you didn't touch.

## About the data

- **Source:** daily visibility for customer brands tracked in 3RD, calculated the same way as the dashboard, 5 August – 2 November 2026.
- **Brands:** 151 customer brands with at least 35 days of 30 or more answers. Days with fewer than 30 answers were left out. Internal, demo and test workspaces are excluded.
- **Weekly moves:** the absolute change between consecutive 7-day averages. 1,276 weekly changes in total.
- **Volume groups:** brands split into three equal groups by average answers per day.
- **Caveats:** we can't see every change a customer made in the period, so some of the movement may come from real work. That makes the "normal" bands, if anything, a little wide. Customers choose their own prompts, and the sample leans Nordic. We report medians across brands unless we say otherwise, and every number is aggregated and anonymized.

## FAQ

**How much does AI visibility normally change week to week?** In our data, the median weekly move was 1.1 points, and 90% of weekly moves were under 4.3 points.

**Why did my AI visibility drop overnight?** Most often because answers vary from day to day, or because an assistant changed models. Look at the weekly average before you act.

**How long until a change I made shows up?** Days to weeks. AI assistants re-read the web on their own schedule, so judge your work on the trend over a month.

**Do bigger brands have more stable visibility?** Brands with more answers per day have steadier numbers. It's about sample size, not brand size.

**What counts as a meaningful change?** A move of 4–5 points or more that holds for two weeks, or a steady drift in one direction over a month.

## Next step

This week, open your last 12 weeks of visibility and note the biggest weekly move. That's your normal band. Put it next to the number in your next report.

If you want the trend, the model changes and the cause in one place, [start for free](https://app.get3rd.com/onboarding/signup) and switch the Overview to the last 90 days.