What an SEO forecast can and cannot promise
A forecast is a model of the future built from assumptions. For organic search those assumptions are unusually uncertain: positions depend on how a search engine assesses the site against every other page competing for the query, results pages change, demand moves with seasons and markets, and competitors publish too. A forecast that hides this behind one precise number is a promise nobody can keep. One that states its assumptions and gives a range is a plan that can be discussed, funded and checked.
The SEO KPIs and reporting guide treats the forecast as the sum of per-cluster targets and asks for the assumptions to be listed. This guide builds that forecast end to end. It has five components, and keeping them separate is what makes the forecast checkable later.
| Component | Question it answers | Main data source | Main uncertainty |
|---|---|---|---|
| Baseline | What will existing pages earn if no new work is done? | The site's own Search Console and analytics history | Whether the trend and seasonal pattern continue |
| New opportunity | What could planned clusters add? | Impressions the site already earns for those queries; keyword tool estimates | Whether pages reach the positions assumed |
| Click-through | How many clicks does each position bring? | The site's own query data by position band | Result features, brand and device mix |
| Timing | When do gains arrive? | The site's record with earlier launches | Release dates, competition, how quickly pages are found |
| Outcomes | What do the clicks produce? | Analytics conversion rates by page type; value per conversion | Conversion rate, lag and attribution |
What a forecast can honestly promise is narrower than most stakeholders want and more useful than they expect: the likely range of outcomes if the plan is delivered, the assumptions that move that range most, and a monthly check that tells everyone early whether the plan is on track. The SEO measurement hub sets out what organic search data can and cannot prove, which is the frame for every assumption below.
The baseline: trend and seasonality from your own data
The baseline is the forecast for pages that already exist, assuming the site keeps them in their current state. It usually carries most of the forecast's volume, so errors here matter more than errors in new clusters.
Export the history. The Search Console (GSCGoogle's search performance and indexing tool.) performance report gives clicks, impressions, click-through rate and average position by query, page, country, device and date, and keeps sixteen months of data, so export it regularly or connect the bulk export if you need longer series. The Search Console tutorial covers filters, exports and the common misreadings. Analytics organic sessions can supply a longer seasonal history; they count a different event, so use them for the shape of the year, not for the level of clicks.
Segment before you project. Brand queries follow brand activity such as advertising, press and word of mouth, not the SEO programme, so forecast them separately or hold them flat as a stated assumption. Project non-brand clicks by page type or cluster, and by country or device where the site serves more than one, because each segment has its own trend.
Measure seasonality. For each segment, divide each month's clicks by the average month for its year to get a seasonal index. One full year gives one observation of the pattern; two or more years show whether it repeats. Where the history covers only a single cycle, say so: a one-year seasonal index treats every quirk of that year as normal. Search demand data from a keyword tool can confirm the shape of the year for the category (SeasonalityPredictable, repeating changes in demand tied to the calendar, such as months, holidays, weekdays or hours.).
Measure the trend. Compare each month with the same month a year earlier, which removes most of the seasonal effect, and read the direction across the year. Then decide, and write down, whether the trend continues, flattens or reverses. Older content often loses clicks slowly as competitors publish and results pages change; the site's own record for pages that were not refreshed shows how fast.
Remove one-off events. A migration, an outage, a tracking change, a page that briefly went viral or a campaign that drove brand searches all distort the history. The annotations log the KPI guide recommends is where these are found. Exclude or adjust those months rather than projecting them forward.
The baseline for each month is then the segment's underlying level, adjusted by the trend assumption and multiplied by that month's seasonal index. Keep the arithmetic in a sheet where every input sits in its own cell with its source beside it.
Sizing new opportunity
New opportunity is the extra clicks planned work could add: new pages for clusters the site does not yet cover, improved pages for clusters where it ranks too low, and technical fixes that let existing pages compete. Work cluster by cluster from the keyword map, with the query set, the target page and the current position for each.
Demand from your own data first. Where the site already appears for a cluster's queries, even at low positions, Search Console impressions show real searches for which the site was shown. Treat them as a floor rather than the full demand, since an impression requires a search but not every search produces one for your site.
Keyword tool estimates second. Where the site has no presence, use keyword tool estimates of Search VolumeThe estimated number of times a query is searched in a period, usually per month., and treat them as estimates. A useful check is calibration: for queries where the site already ranks near the top and earns impressions, compare the tool's figure with the impressions Search Console records. If the tool consistently reads higher or lower than your own data, apply that difference to its figures for new clusters and record it as an assumption.
A realistic target position. Review the results page for each cluster's main queries: who ranks, what kind of page wins, how much space ads and features take. Choose the position the site can plausibly hold given its strength in the topic, and apply the same standard to every cluster. The keyword prioritisation page for the SEO traffic estimator uses the same inputs to rank candidate clusters against each other; the forecast then takes the chosen clusters forward in time.
Net, not gross. A new page can take clicks from an existing one targeting the same queries, and a cluster already partly ranking contributes clicks to the baseline. Count only the increase over what the baseline already includes, or the forecast counts the same clicks twice.
Click-through assumptions from your own data
Click-through rate (CTRShare of impressions that result in a click.) converts impressions at a position into clicks, and it is where forecasts most often borrow numbers they should not. A published curve averages sites, query types, devices and results pages that are not yours. The share of clicks a position earns changes with the ads and features above it, with whether searchers know the brand, with intent, and with device. Your own data describes the results pages your site actually competes on.
To build your own click-through table:
- Export query-level data from the performance report for a recent period of several months: query, page, clicks, impressions and average position, split by device.
- Exclude brand queries, which behave differently from non-brand ones and would distort every band.
- Keep queries with enough impressions to give a stable rate, and set that threshold once.
- Group the remaining queries into position bands by their average position.
- For each band, divide total clicks by total impressions. That is your click-through for the band.
- Where the data allows, split the table by page type or intent, and by device, because each has its own pattern.
Read the result with the report's definitions in mind. Average positionMean ranking of a site's topmost result across the impressions it received. is the mean of the topmost position the site held across impressions, so a query averaging a middle position may be alternating between a high and a low one; shorter periods and device splits reduce that blur. The query table also leaves out anonymised queries, so query-level totals sit below the property's chart totals; use the property totals for the baseline and the query table for the click-through table.
| Data choice | Why it matters for the forecast |
|---|---|
| Brand queries excluded | Brand click-through would inflate every band |
| Split by device | Mobile and desktop results pages differ in layout and click behaviour |
| Split by page type or intent | Product, category and guide queries attract different result features |
| Minimum impressions per query | Small samples produce rates that swing from month to month |
| Search appearance noted | Rich results and features change click-through at the same position |
| Period of several months | Smooths weekly noise without mixing very different seasons |
Where the site has never ranked in a band, for example at the very top for a competitive cluster, there is no own data for it. Borrow the rate from the nearest band you do have, adjust it with a stated reason, and list it as the weakest assumption in the forecast. The CTR calculator is useful for checking the click-through implied by any pair of click and impression figures, and the SEO traffic estimator shows the arithmetic of volume, position and click-through, but its curve is illustrative: replace it with your own table before the numbers go into a forecast.
Timing and ramp-up
Gains do not arrive when the work is approved. They arrive after the page is written and published or the fix is released, after the search engine finds and processes the change, and after the page earns its position, which can take a long time for competitive clusters. A forecast that books the full gain in the month of publication will look like a failure for months and then overshoot.
Take timing from the site's own record. For pages published over the past year or two, find the first month with impressions and the month each reached a stable position, using the page filter in the performance report. The typical gap between publication and stable position, and the shape of the climb, become the ramp-up assumption for similar pages. Technical fixes and template changes start from their release date, which depends on development capacity, so ask the team that owns the release queue rather than assuming the next sprint.
Seasonality interacts with timing. A cluster launched just after its seasonal peak shows little until the next season, however well it ranks, so apply the seasonal index to the new clusters as well as the baseline.
Scenarios and ranges
A single forecast line hides the uncertainty the plan carries. Three scenarios show it: a low case that is plausibly disappointing, a central case that reflects the most likely assumptions, and a high case that is plausibly good. None is a worst or best imaginable case. Vary the assumptions that carry most uncertainty and leave the well-measured ones alone.
| Assumption | Low scenario | Central scenario | High scenario |
|---|---|---|---|
| Baseline trend | The decline seen on unrefreshed pages continues | The recent year-on-year trend holds | The trend improves where refresh work is planned |
| Positions reached by new clusters | One band below the target | The target position | One band above the target for the strongest clusters |
| Click-through | Own rate for the band, adjusted down where features are expanding | Own rate for the band | Own rate for the band |
| Timing | The slowest ramp-up in the site's own record; releases slip | The typical ramp-up; releases as planned | The faster end of the site's record |
| Conversion | The lower end of the page type's recent range | The page type's recent average | The page type's recent average |
Before finalising, test sensitivity: change one assumption at a time across its plausible range and see how much the total moves. The assumption that moves it most is the one to monitor first and to improve with better data. A KPI tree helps here, because it shows which input sits behind each outcome. The marginal returns and forecasting guide sets out the same scenario logic for paid spend; the difference for organic search is that costs come first and results arrive later, so the scenarios differ more in timing than in cost. The academy exercise on forecasting a dog food maker's breed page programme practises building low, central and high cases from the site's own baseline and click-through before the programme is approved.
From clicks to outcomes
Clicks are a middle link in the chain. The forecast that a budget holder needs is in outcomes: orders, revenue, leads or sign-ups.
- Clicks to sessions. Search Console clicks and analytics sessions count different events, so they never match. Measure the ratio between them on organic landing pages for a recent period and apply it, rather than treating one as the other. The comparison of Search Console and GA4 explains why they differ.
- Sessions to conversions. Use the conversion rate for the page type or intent from your own analytics, not the site average: guides convert differently from category pages. New clusters take the rate of the most similar existing pages.
- Conversions to value. Apply the average value per order or the measured lead-to-customer rate and customer value, from the business's own records.
- Lag. Where buying takes time, outcomes arrive after the visit. Spread them across later months using the lag the business observes.
- Attribution. State the model used to credit organic search, as the KPI guide asks, so the forecast and the later report use the same rules.
Each step is a stated assumption with a source. The output is a table of outcomes per month for each scenario, with the baseline and each new cluster shown separately, so that each line can be checked later (KPIA metric tied directly to an objective.). Turning those scenarios into a funding decision, with full costs, payback periods and staged reviews, is the job of the guide to building an SEO business case.
Tracking forecast against actual
A forecast earns trust by being checked. Each month, set the actual results beside the forecast, line by line, and classify any gap.
| Line | What to compare | A gap usually means |
|---|---|---|
| Baseline | Actual non-brand clicks to existing pages against the baseline forecast | The trend or seasonal assumption was wrong, or a one-off event occurred |
| Each new cluster: timing | When the page went live and first earned impressions, against plan | Release slipped or discovery was slow |
| Each new cluster: position | The position reached against the target | The position assumption was optimistic or competition changed |
| Each new cluster: click-through | Clicks per impression at the reached position against the table | Result features or snippet quality differ from the assumption |
| Outcomes | Conversions and value against forecast | Conversion rate, lag or attribution assumption was wrong |
Splitting gaps this way follows the volume, rate and timing logic in the missed marketing forecast diagnosis and stops the review from arguing about the total. Revise the forecast quarterly, keep each version, and explain each revision in the report. Feed what you learn back into the method: each launch adds a data point to the ramp-up record, and each quarter of query data refreshes the click-through table, so the next forecast rests on more of the site's own evidence and less assumption.
A bicycle retailer forecasts a year of organic clicks
Fiddlecrest Bicycles, an invented online bicycle retailer, forecasts non-brand organic clicks for its next financial year. Its baseline uses two years of history from Search Console exports and analytics. Non-brand clicks to existing pages were 210,000 in the last twelve months; the year-on-year trend on pages it has not refreshed is about 3% down, which the central case keeps. The seasonal index runs from 0.6 in December to 1.5 in May. Three clusters are planned: a gravel bike category, an e-bike commuting guide set and a bike sizing guide. Fiddlecrest's own click-through table for non-brand queries, averaged across devices for this summary, reads 16% for positions one to three, 7% for four to six and 3% for seven to ten. Its pages published over the previous year took about five months to reach a stable position, so each cluster ramps in over five months from its publication date. The scenarios give 202,000 clicks in the low case, 232,000 in the central case and 258,000 in the high case, with the baseline contributing about 204,000 of the central figure. Applying its own conversion rates by page type, from 1.4% on category pages to 0.3% on guides, the central case implies about 1,900 orders from non-brand organic search. After the first quarter, the tracking table reads:
| Line | Central forecast, Q1 | Actual, Q1 | Gap type |
|---|---|---|---|
| Baseline, existing pages | 44,000 | 45,100 | Within range |
| Gravel bike category | 1,200 | 300 | Timing: launched seven weeks late |
| E-bike commuting guides | 900 | 1,050 | Position one band above target |
| Bike sizing guide | 400 | 160 | Click-through: a result feature answers the query |
The team moves the gravel category's ramp-up back by two months, lowers the sizing guide's click-through assumption to the rate it actually earns, and reissues the forecast as version two.
Common mistakes
- Borrowing a published click-through curve. It describes other sites' results pages. Build your own table from Search Console, and label any borrowed band as an assumption.
- Forecasting totals. Brand and non-brand, and different page types, follow different trends; project them separately.
- Projecting a one-off. A migration dip or a viral spike in the history becomes a false trend.
- Booking gains in the month of publication. Use the site's own ramp-up record.
- Counting the same clicks twice. New pages that take clicks from existing ones add only the difference.
- One number instead of a range. A single figure becomes a promise; scenarios show the uncertainty.
- Clicks presented as sessions or revenue. Each conversion step needs its own measured rate.
- Never checking the forecast. Without a monthly forecast-versus-actual table, the same wrong assumptions return in next year's plan.
