SEO Forecast
Free, no signup. Most traffic forecasts treat an algorithm update like noise that will smooth itself out — it won't. Upload your Search Console, GA4, Ahrefs, Semrush, or Adobe export and get a forecast that treats updates as the permanent level shifts they are, backtested on your own history before you're asked to trust it.
Runs in your browser — uploads are not sent to our server, but the current project is autosaved in this browser’s local storage until you replace it or clear this site’s browser data. Creating a share link is opt-in; it uploads the scenario you choose and retains it for 7 days. Anonymous run-level outcome counters may be used for aggregate research; URLs, domains, IPs, and identifiers are never included, and no statistic is released below 100 runs.
Google algorithm-update data last synced Jul 7, 2026.
Sample data. This is a generated demo series (with a planted algorithm-update hit and a fake migration) so you can see how the tool works. Upload your own export to replace it.
Detected events — tell the model what happened
Level shifts and outliers found in your data. Label them so the model treats them correctly; leave as "ignore" to keep the data untouched. Known Google updates are handled automatically and only kept when they measurably moved your series.
How wrong was each model on your own history?
Rolling-origin backtest: each model repeatedly forecast a held-out slice of your data it had never seen. MASE < 1 means it beat the seasonal-naive baseline. The chart overlays what the top models would have predicted for your most recent data.
Forecast every series you've loaded and compare them on one chart, or as share of voice. Mark which are you vs competitors — needed for calibration and the crossover callouts. Add competitors by uploading more files above (wide files with a column per site, or one file each).
Calibrate competitor estimates to your real data
Third-party traffic estimates (Ahrefs/Semrush) are directionally useful but wrong in absolute terms. If you upload your organic ground truth (GSC clicks or a GA4 organic segment) plus your series from the same third-party tool, we compute your correction factor and apply it to competitors.
Forecast the traffic from ranking changes — no time series needed. Enter keywords with their search volume, where you rank now, and where you're aiming. This is scenario math (volume × click-through rate by position), not curve-fitting.
Keywords (0) — edit any cell to update the forecast
Incremental clicks = the extra monthly clicks at the target position vs. today, probability-weighted.
Business case
Turns the forecast bands into a conversion, revenue, ROI, and paid-search-equivalent view. These are planning assumptions, not a revenue prediction.
Forecasts assume no future algorithm updates, content changes, or SERP-feature shifts — the intervals come from each model's real backtest errors on your data, not from optimistic formulas. Third-party traffic and traffic-value figures are estimates. Nothing you upload is sent anywhere: parsing, modeling, and charts all run locally in your browser.
Sample report
Suppose you upload a weekly Search Console clicks export that dips after the March 2025 core update and recovers late in the year:
Date,Clicks,Impressions,CTR,Position
2025-01-06,9800,410000,2.4%,7.1
2025-01-13,9950,415000,2.4%,7.0
2025-03-17,8200,398000,2.1%,8.3
2025-06-02,8600,405000,2.1%,8.0
2025-11-10,10400,430000,2.4%,6.6…and the Forecast tab returns a summary like this (illustrative projection — invented numbers, not real data):
Illustrative example — forecasting your own export runs entirely in your browser; the numbers below are fabricated to show the summary's shape, not a real backtest
- The 80% range is the actual forecast, not the 11.6k headline number — a spread of 9.8k–13.5k means real outcomes anywhere in that band are consistent with the model. See how to read the range ↓.
- The model name and MASE score explain why you should trust it — Theta won because it had the lowest backtest error; MASE below 1 means it beat a naive seasonal-repeat baseline, not an arbitrary bar.
- Only algorithm updates that measurably moved this series appear — the March 2025 core update is listed because the model detected an actual ~16% shift, not because of when it shipped. The methodology behind this detection is covered in SEO forecasting.
How to use it
- Upload a CSV (or click Load demo data to explore first). Search Console, GA4, Ahrefs, Semrush, Adobe, Google Trends, and generic date-plus-value files are auto-detected — you don't pick a format.
- Pick the series to forecast (clicks, sessions, position, traffic value…), set group by to weekly or monthly, and drag the horizon slider to how many periods ahead you want.
- In Detected events, label any level shift the tool found — a migration, redesign, or tracking change — so it's modeled correctly. Known Google updates are handled automatically.
- Leave model on auto (best backtest) or force a specific one, then press Forecast.
- Read the summary, then switch tabs: Explain for the seasonal-vs-real verdict and per-update impact, Backtest for model accuracy, Keywords for a ranking-change scenario. Export the forecast as CSV or PNG.
SEO Forecast projects organic traffic from your own analytics export and treats Google algorithm updates as permanent level shifts rather than temporary blips. For the theory behind the approach see SEO forecasting; to turn a projected traffic gain into a business case, pair it with the SEO ROI Calculator.
How to read the estimate
- The big % + arrow — projected change from now to the end of the horizon. Green is favourable, red unfavourable (and for inverted metrics like average position, a lower number is the good direction).
- 80% / 95% bands — prediction intervals from each model's real backtest errors. The 80% band is your realistic planning range; it widens further out because uncertainty compounds.
- Central dashed line — the model's best single estimate; the solid line is your actual history.
- Update markers — a bold vertical line with a measured % means a Google update moved your series; a faint line means a known update had no measurable effect here and was left out of the math.
- Anomaly dots — one-off outliers (an outage, a tracking glitch) that were down-weighted so they don't distort the baseline.
- Data quality score — a 0–100 read on whether you have enough clean history (ideally two-plus yearly cycles) for seasonality to be identifiable.
- MASE in the backtest table — mean absolute scaled error; below 1 beats the seasonal-naive baseline, and lower is better.
How it works
Everything runs client-side. Your CSV is parsed and auto-format-detected, then the heavy work — decomposition, level-shift and outlier detection, model fitting, and a rolling-origin backtest — happens in a Web Worker so the page stays responsive and your data never leaves the browser. The tool fits several models (seasonal-naive, drift, ETS, Theta, and an updates-aware curve fit), scores each by repeatedly forecasting held-out slices of your own history, and picks the winner. Google updates from a bundled dates file are tested as candidate level shifts and only kept when they measurably moved your series. Prediction intervals come from those backtest errors, not from a formula — which is why they're honest about uncertainty.
Features
- Auto-detects GSC, GA4, Ahrefs (traffic + traffic value), Semrush, Adobe, Google Trends, and generic exports; weekly/monthly/as-uploaded aggregation.
- Five backtested models with an auto-pick by lowest MASE, plus a manual override.
- Algorithm updates modeled as level shifts, kept only when they moved your data; unlabeled shifts surfaced for you to label.
- Explain tab: seasonal-vs-real drop verdict, trend/seasonal decomposition, and counterfactual per-update impact.
- Backtest tab: how wrong each model was on your history, with an overlay of what the top models would have predicted.
- Compare & SoV tab: forecast multiple sites on one chart or as share of voice, with competitor-estimate calibration to your ground truth.
- Keywords tab: scenario math (volume × CTR-by-position) for ranking changes — no time series needed, with optional CTR-curve fitting from your GSC Queries export.
- Stress-test overlay, 80/95% bands, log scale, saved-scenario ghost line, and CSV/PNG/scenario exports.
Limitations
A forecast assumes no future algorithm update, content change, or SERP-feature shift during the horizon — it can't predict events that haven't happened, which is exactly why the intervals widen. It needs enough clean history (roughly two yearly cycles) for seasonality to be reliable; short or gappy series get wide bands and a low quality score. Third-party traffic and traffic-value figures (Ahrefs/Semrush) are estimates, not measured clicks — calibrate them to your own ground truth before trusting absolute numbers. The Keywords tab is scenario math, not curve-fitting, so it's only as good as the CTR curve and hit-probability you give it. And the example figures on this page are illustrative, not a benchmark for your site.
Frequently asked questions
Can you really forecast SEO traffic accurately?
No forecastSEO forecasting uses historical data — traffic, click-through rate, rankings, and search volume — to project future organic search performance and its business impact. It's a probabilistic model under defined assumptions, not a guarantee of results. is exact, and this tool is deliberate about saying so. Instead of a single confident line, it fits several models, backtests each one on slices of your own history it never saw, and reports a prediction interval built from those real errors — not from an optimistic formula. The forecast is only as good as the assumption that no future algorithm update, content change, or SERP-feature shift lands during the horizon, which is why the intervals widen the further out you look. Treat the central line as a planning midpoint and the 80% band as the realistic range.
How does the tool handle Google algorithm updates?
Known Google updates from a bundled dates list are tested against your series as candidate level shifts — permanent step changes in the baseline rather than temporary dips. An update is only kept in the model when it measurably moved your data; if a core update did not touch your site, it is shown faintly and left out of the math. The tool also detects unlabeled level shifts (a migration, a redesign, a tracking change) and asks you to label them so they are modeled correctly instead of being smoothed away.
What CSV formats can I upload?
Search Console, GA4, Ahrefs (traffic and traffic value), Semrush, Adobe Analytics, Google Trends, and a generic date-plus-value layout are all auto-detected — you do not pick a format. It reads any file with a recognizable date column and one or more numeric columns, so you can forecast clicks, sessions, impressions, average position, or estimated traffic value. Everything is parsed in your browser; no file is uploaded to a server.
Is my data uploaded to a server?
No. Parsing, model fitting, backtesting, and chart rendering all run locally in your browser, with the heavy computation in a Web Worker so the page stays responsive. Your export never leaves your machine. The only thing that can be sent anywhere is a share link, and only if you explicitly click Create share link.
What is a prediction interval and why does it widen?
A prediction interval is the range the tool expects the true value to fall in, at a stated confidence — the 80% band should contain the actual outcome roughly 80% of the time. It comes from how wrong each model was during backtesting on your history. The band widens further into the future because uncertainty compounds: a small weekly error accumulates over many weeks, and the chance of an unmodeled event landing grows the longer the horizon.
Site passport Local context for this saved site
Local data
Saved targets, named lists, and recent check summaries remain only in this browser.
Rate this tool
Feature requests for Seo Forecast
Upvote what you want most. New ideas can be submitted from the floating Feedback menu; requests appear here once approved, and the most-wanted rise to the top.
You won't be emailed about that request anymore.
Loading…
➕ Request a feature
New requests are reviewed before they appear here.
Common issues & how to fix them
- Warnings Traffic has a lasting level shift Fix: Identify the deployment, migration, algorithm event, or tracking change at the shift date and model pre- and post-shift periods separately after correcting the cause.
- Errors Labeled event has a negative estimated impact Fix: Segment the affected pages and queries around the labeled event, reverse or repair the causal change where evidence supports it, and measure against an unaffected comparison segment.
- Warnings History is too short for a stable forecast Fix: Provide at least one full seasonal cycle of consistent-period data when possible, or shorten the forecast horizon and label the estimate as unstable.
- Information Forecast is strongly seasonal Fix: Compare the same seasonal period year over year and plan against the seasonal baseline instead of the latest point alone.
- Warnings Forecast uncertainty interval is wide Fix: Shorten the forecast horizon or add cleaner history before using the midpoint for planning.
- Warnings Backtest accuracy is weak Fix: Use the prediction interval, test a baseline model, and collect more consistent history before relying on the forecast.