What a Forecast Is For
A traffic forecast answers a planning question: if we publish these pages, what range of results should we expect, and by when? It is not a promise. A useful forecast states its assumptions, gives a range, and can be checked later against what happened.
Step 1: Build the Baseline
Export monthly clicks and impressions from Search Console for as long as you have data. Mark anything unusual: site changes, a migration, a spike that came and went. One-off days should come out of the baseline. We exclude a single day in our own data where clicks jumped to several times normal with no change in impressions and fell back within two days.
Step 2: Separate Old Pages From New
Existing pages and new pages behave differently. Forecast them apart.
- Existing pages: project the recent trend, adjusted for season if you have more than a year of data.
- New pages: model them as cohorts by the month they were published.
Step 3: Model New Pages as a Funnel
For a planned batch, traffic is a chain of four numbers.
| Step | Question | Where the number comes from |
|---|---|---|
| Pages published | How many, and when? | Your plan |
| Share indexed | What fraction does Google index? | Your earlier batches, in Search Console |
| Share earning impressions | What fraction shows up for any search? | Your earlier batches |
| Clicks per page earning impressions | How many clicks does a typical such page get per month? | Your earlier batches, using the median, not the average |
Use the median for clicks per page. A few pages usually earn most of the traffic, and an average built on them overstates what a typical new page will do. On this site, one section holds 579 of our top 1,000 pages and earns 91% of the clicks to them.
Step 4: Add the Time Curve
New pages do not reach their level at once. Look at how long earlier batches took to be indexed and to settle, and apply the same delay. Our own site took about three months to move from tens of clicks a month to hundreds.
| Month, 2026 | Google clicks on presenc.ai |
|---|---|
| March (from the 16th) | 20 |
| April | 15 |
| May | 125 |
| June | 380 |
| July | 840 |
| August | 1,390 |
| September | 2,100 |
| October | 2,860 |
Figures are rounded. This is one site in one subject. Use your own history where you have it.
Step 5: Give a Range
Run the funnel three times: with cautious, expected and optimistic values for each step. Report the three results. The range should widen the further out you go, because each month's error carries into the next. A single number hides how uncertain it is.
Step 6: Test It Against Months It Never Saw
Before trusting the method, hide the last three months of your data, forecast them from what came before, and compare. If the real numbers fall outside your range, the range is too narrow or an assumption is wrong. This is the same idea as holdout validation in budget modelling.
Step 7: Forecast the Outcome, Not Only the Clicks
Clicks are a step toward something. Multiply through by what you measure: the share of visitors who sign up, enquire or buy from pages of this kind. If you have no such figure for a new page type, say so, and treat the first batch as the measurement.
Step 8: Write Down the Assumptions and Revisit
Record what you assumed for each step and when. Compare monthly. The comparison tells you which assumption was off, which is more useful than knowing the total was wrong.
What a Forecast Cannot See
- Search engine changes. Ranking updates and new result formats can move traffic in either direction.
- Answers without clicks. AI answers and summaries can satisfy a search without a visit. See zero-click attribution.
- Competitors. Someone else may publish a better page.
- A new subject. History from one topic may not carry to another.
Forecasting Across Channels
Organic search is one channel. Forecasting total sales, or the effect of moving budget between channels, needs a model of all of them. See marketing mix modeling and how to calibrate a model with lift tests.
If You Want This Done for You
Presenc AI builds search pages with AI agents today, with a person reviewing every page. Forecasting and budget modelling agents are in development. Request access.
Sources
Monthly clicks are from Google Search Console for presenc.ai. The method is general forecasting practice applied to search data, and the worked figures describe this site only.