SEO Forecasting: The Assumptions a Tool Cannot Make For You

Nothing in the forecast was wrong. That is what makes it hard to argue with. The traffic landed where the spreadsheet said it would, the board slide went out on time, and the revenue that traffic was supposed to produce arrived at a fraction of the number in the deck. Nobody had lied. Somewhere between a row of cells and a promise about next year, a handful of multipliers had been filled in, and the person who could explain them had left the company.

The arithmetic was never the problem. An SEO forecast is a list of queries multiplied by a click-through rate, shaped by a ramp, filtered by the share you expect to win, and reduced by whatever the new pages take from the old ones. Software will do the multiplying. It will not tell you that the four multipliers underneath it are somebody else's borrowed guess.

You can buy a better multiplication table. You cannot buy your way out of choosing the multipliers.

What an SEO Forecast Is Actually Made Of

Every organic forecast resolves to the same object: a set of queries with an assumed demand. A click-through rate assigned to the position you expect to hold. A curve describing when that traffic arrives. A share of the set you believe you can win. A subtraction for the pages that already rank for the same things.

The software handles the second half: per-URL aggregation, the scenario switch, the chart your client will screenshot. The first half is yours.

Two teams can drop the same domain into the same tool on the same afternoon and come back with numbers that are not close. One assumed position three and a six-month ramp; the other assumed position one and a year.

Which outputs your business cares about is a separate question, covered in marketing KPIs, and a forecast sits inside that work, one layer within the frame described in marketing measurement.

Assumption One, Click-Through Rate by Position

Borrowing a curve feels responsible. It is the opposite. Four credible studies of the same quantity disagree by an order of magnitude, all four published or updated in the last eighteen months.

• First Page Sage's 2026 report, which states no sample, puts organic position one at 22.6 percent click-through when no AI answer sits on the page, and 3.6 percent when one does.
• Seer Interactive, publishing November 4 2025 from 3,119 queries across 42 organisations, measured organic click-through on AI Overview queries falling from 1.76 percent to 0.61 percent between June 2024 and September 2025.
• Pew Research Center, publishing July 22 2025 from 900 US adults and 68,879 searches logged in March 2025, found a result was clicked in 8 percent of visits where a summary appeared, against 15 percent where none did.
• Ahrefs, in a February 4 2026 update across 300,000 desktop keywords, reports position one falling from 7.3 percent to 1.6 percent between December 2023 and December 2025.

None of those is wrong. Each answers a different question: a mixed keyword set, a set of AI Overview queries, a controlled browsing study, a desktop sample. Take the newest and apply it to your list and you have borrowed somebody else's search results page. All four were read through a compiled index printing each study's date, window and sample.

The fix is unglamorous and takes an afternoon. Search Console reports position, impressions and clicks per query, so a click-through rate per query is arithmetic on your own export. Fit the curve there, then state the sample, because a curve fitted on forty queries is a curve fitted on forty queries.

Assumption Two, the Ramp Curve

Advanced Web Ranking's template, dated February 25 2025, tells the reader to multiply the traffic estimate by 0.2 percent in month one, 0.3 percent in month two and 0.6 percent in month three, and gives no basis and no sample for any of the three. The sentence after them says only that the percentages reflect slow initial gains. A free calculator on isocialweb.agency, read on October 29 2026, says little moves in the first two months and then keywords close to the top start climbing. That one is unsourced too, and shaped differently.

In a summary of an Ahrefs ranking study whose date and sample the summary does not state, 1.74 percent of newly published pages reach the top ten within a year, against 5.7 percent in 2017. Of the pages that did reach the top ten, 40.82 percent got there within a month, and 72.9 percent of the pages holding a top-ten position today are more than three years old. The primary page returned not found this week, so treat those figures as summarised rather than freshly verified.

Read them together and the shape is a step, not a line. Most pages never arrive. A minority arrive fast, and the ones that do keep growing for years.

A straight line fails in both directions. It overstates year one, assuming every planned page arrives on schedule, and understates year two, assuming the ones that arrived stopped growing. No study I found gives a defensible month-by-month ramp for a planned page on a specific site, so any page handing you a monthly percentage without a sample is guessing in your name. Backtest your own instead: take pages you published eighteen to twenty-four months ago, align them at their publication month, and plot the median share of their settled traffic by month.

Assumption Three, Share of Queries Won

The query list is an estimate before you apply anything to it. Ahrefs' 2021 study of more than 72,000 keywords compared Keyword Planner demand estimates against Search Console impressions: it drastically overestimated 54 percent of the time, was roughly accurate 45 percent of the time, and overstated impressions by at least four times on 14.8 percent of keywords. Ahrefs was grading a competitor, so read the comparison with that in mind; the absolute finding is the part worth carrying.

Then comes the click problem. First Page Sage's 2026 report, which states no sample, puts 74.2 percent of US searches ending without a click at all. Demand that never becomes a session cannot become revenue, however well you rank for it.

Before any of that, deduplicate. Two spellings of one thing are two rows in an export and one demand, and the pair you typed to find this page is exactly that shape. Singular, plural and reordered variants arrive by construction in anything a tool hands you. Collapse them first, or you are counting the same person four times.

Then make three subtractions in this order: what you cannot rank for, what does not convert, and what you already hold. Only the middle one has hard external evidence. The other two are judgement, and judgement is easier to defend itemised.

I could not find a study giving a defensible average share of queries won by a new content programme. Any percentage on that is an assumption, including the one in your last deck. Set it from your own record instead: of the pages you published in the last two years, what share hold any top-ten position at all?

Assumption Four, Cannibalization and Subtraction

Studio 36 Digital's keyword cannibalisation study, analysis dated January 14 2026, covered 2,500 keywords, the top 25 on each of 100 high-authority sites across ten industries, measured on google.co.uk. The average successful site in that 100-site UK sample ranked 4.7 URLs for every keyword. 68 percent of those 100 sites ranked five or more URLs for a keyword, and only 12 percent of them kept it to one or two. The extreme case, in the same 2,500 keywords, was 59 URLs on a single query.

It is a UK dataset, and most readers do not operate in that market. The pattern is plausible elsewhere; the counts are UK counts.

The study's own conclusion cuts against the easy reading. Its authors argue that overlap is normal and high-authority sites live with it. That does not weaken the subtraction step, it sharpens it. What it kills is the idea that overlap is a defect. A new page joins a stack that already exists, so the subtraction is about the difference it makes, not about reaching zero.

If a new page ranks for a query an existing page already holds, part of the click gain is a transfer rather than a gain. The net is smaller than the gross, and a forecast that ignores the difference overstates from the first cell.

You can measure it instead of assuming it. Record the cluster's baseline before publishing, then after 60 to 90 days compare the cluster total with the new page's own clicks. The gap is the subtraction, measured on your data. Deciding which page should own which query is a planning question with its own home, in SEO and content strategy.

The Three Sanity Checks

Does the implied traffic exceed every click that exists for that query set?

If the forecast implies you capture more clicks than exist, the share assumption is wrong. Add up the demand, apply your share, compare the result with the forecast. Ten minutes, and it catches errors that survive a polished review.

Does the forecast survive a 30 percent haircut on every assumption at once?

Not one at a time, which is what a scenario switch gives you. All four, in the same direction, at the same time. If the plan only works at the optimistic end of four independent guesses, you are not forecasting. You are hoping with a spreadsheet.

Does the model predict any page it did not plan?

If every page in the forecast is a page you intend to publish, the model is describing your plan, not your site. Real sites gain traffic from pages nobody planned: a comparison page, a support article, a glossary entry. A model that cannot hold a page you did not plan will understate consistently.

Keep the model maintained by putting last quarter's forecast next to what happened and writing one line on which assumption broke. That is maintenance, not reporting, and the comparison panel belongs to your marketing dashboard. Change one assumption at a time and keep the previous versions.

Where SEO Forecasting Tools Fit and Where They Do Not

Tools are good at exactly the half of this that is solvable: multiplying, aggregating per URL, redrawing the chart when you edit a cell. Paying more buys interface, and interface does not improve a number. The page holding the top editorial slot for this query is a method guide, not a shopping list.

Tools are also not uniformly silent on the assumptions. A free browser tool published on Patrick Stox's site under /tools/seo-forecast/, read on October 29 2026, fits a click-through curve from your own Search Console export, takes a ramp in months, and reports a prediction interval instead of one headline number. Its own copy says the 80 percent range is the forecast. That is two of the four done properly, for nothing.

What it cannot do is tell you which of your queries you can win, or what a new page will take from an old one. Those two stay judgement, and judgement does not come in a subscription.

So the criterion is narrow: pick the tool that lets you see and edit every assumption, and prints the sample behind each one. Connectors are beside the point. A connector moves your data around; it does not improve your guess. When a vendor's own template tells you to estimate the ranking input with "forecasting tools like" its product, as Advanced Web Ranking's does, you are watching an assumption being sourced from the thing being sold.

Conclusion

Nobody lied in the projection that started this piece. Four numbers went unrecorded, and the projection was their sum. Traffic was the easy part: one multiplier, visible and defended in a weekly meeting. Revenue sat three multipliers downstream, and by the time anyone asked, the person who set them had moved on.

That is what a defensible forecast buys you. Not accuracy, which nobody can promise. A record: this is what we assumed, this is the sample it came from, this is who signed off. When the number misses, you can name the assumption that broke instead of arguing about the tool.

Write the four down before you open the software. Then the harder questions become answerable, because they sit after publication: where the pages actually land, and what the pages already there lose while the new ones climb. Both are measurable, and neither is visible until the assumptions are on paper.