Discussion
  • Agency
  • AI Search

How are you forecasting for new clients now that AI Overviews make CTR-based models far less reliable?

Published 1 reply

Discussion starter: an agency question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.

Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.

1 reply

  • admin
    Admin
    Original poster

    The short answer: stop building new-client forecasts on “search volume × target position × industry CTR curve”. Forecast from the client’s own recent click data, segmented by the kind of results page each query cluster gets, and present ranges tied to leads or revenue rather than one traffic number. The CTR curve still exists, but it now depends on the layout of each results page, and that layout (AI Overview or not, shopping units, local packs) changes faster than keyword tools record it.

    The important qualification is that AI Overviews did not damage forecasting evenly. The published click studies are strongest on informational queries. Commercial, navigational, local and branded queries behave differently. So the fix is not one discount applied across the whole model. It is splitting the model so that each part uses a click rate you have actually observed for that type of query.

    The old model now fails in three separate places

    1. Click-through fell where AI Overviews appear

    Pew Research tracked the browsing of 900 US adults in March 2025. When a Google search showed an AI summary, users clicked a traditional result in 8% of visits, against 15% when no summary appeared, and clicked a link inside the summary in just 1% of visits. About 18% of searches in the sample produced a summary. Ahrefs, using aggregated Search Console data for 300,000 keywords, found the presence of an AI Overview correlated with a 58% lower average CTR for the position one result, comparing December 2025 with December 2023 on desktop. Its April 2025 estimate had been 34.5%.

    Both are correlational. Pew is US-only and a snapshot, and Ahrefs compares periods two years apart on desktop only. They are good reasons to distrust a pre-2024 CTR curve. They are not a discount factor you can paste into a client model, because the effect varies by query type, market and SERP.

    2. The impression baseline shifted in September 2025

    When Google stopped supporting the &num=100 results parameter in September 2025, many sites saw desktop impressions fall and average position improve in Search Console, while clicks were affected far less, according to Brodie Clark’s analysis. Any CTR you calculate from impressions before that change is not comparable with CTR after it. A new client whose CTR “improved” in late 2025 may simply have lost impressions that were inflated by rank-tracking tools. Build CTR baselines from data after the change, and backtest on clicks rather than impressions.

    3. Search Console cannot show you an AI Overview CTR

    AI Mode data has been included in Search Console’s Web totals since June 2025 without a separate click breakdown. The Search generative AI performance reports, which reached all eligible sites on 31 August 2026, show impressions for AI Overviews, AI Mode and AI features in Discover, but no clicks, CTR or queries. Google does not document how to subtract those impressions from Web totals. So you can see AI exposure trends by page, but you cannot compute a clean “click rate when an AI Overview shows” from first-party data.

    There is a fourth, softer problem: rankings and AI citations have drifted apart. Ahrefs reported the share of AI Overview cited pages that also rank in the top 10 falling from 76% to 38% between July 2025 and March 2026, although its detection method changed in between, and BrightEdge reported only 17% overlap using a different method. The measurements disagree, but both say a ranking forecast no longer describes all of a page’s visibility.

    The structure I would use instead

    Non-brand demand by query cluster (12+ months, for seasonality)
    ↓
    Achievable position band for each cluster, given competition and the planned work
    ↓
    Click share for that band on that SERP type (the client’s own observed CTR, post-September 2025)
    ↓
    Conversion rate by landing-page type (not a sitewide average)
    ↓
    Value per conversion, supplied by the client
    ↓
    Conservative, base and upside ranges, each with named assumptions
    1. Get Search Console access during the sales process. Pull the full 16 months via the API. Split branded from non-branded using the branded queries filter or your own regex. Forecast branded clicks separately and conservatively, because SEO work rarely moves brand demand quickly.
    2. Cluster queries by intent and tag the SERP type. Use a rank tracker that records whether an AI Overview, shopping unit or local pack appears for each tracked query. Accept that AI Overview presence changes over time, and re-check it at each forecast review.
    3. Calculate the client’s own CTR by position band for each cluster type, using only data from after the September 2025 change. Where the client has no data for a cluster (new content, new market), borrow the rate from their most similar existing cluster and widen the range.
    4. Calibrate keyword tool volumes. For queries the client already ranks on page one for, compare tool volume with Search Console impressions. The ratio tells you how far to trust tool estimates for the gap keywords in that niche.
    5. Write the assumptions next to the numbers. Each scenario should state its ranking gain, whether AI Overview prevalence in that cluster holds or grows, and whether CTR keeps declining.
    6. Backtest before presenting. Run the same method on data from a year earlier and compare the prediction with actual clicks. The error you see there is the minimum honest error band for the new forecast.

    What changes in the inputs

    Old input Replacement Evidence to inspect
    Industry CTR curve Client’s own CTR by position band and SERP type Search Console by query cluster, post-September 2025 only
    Keyword tool volume Search Console impressions for existing coverage; calibrated tool volume for gaps Ratio of tool volume to impressions on page-one queries
    One traffic number Three scenarios with named assumptions Which assumption moves the forecast most (that is the one to discuss with the client)
    Traffic as the outcome Leads or revenue by page type Analytics conversions by landing page and channel
    Rankings as the only visibility measure AI impressions and citations reported as leading indicators, not revenue Search Console generative AI report; Bing Webmaster Tools AI Performance citations

    Hypothetical example:

    A new B2B software client wants a content programme. The informational cluster (“how to” and definitional queries) has tool volume of about 50,000 searches a month, and most of those queries show an AI Overview. An industry curve at an average position of 3 might suggest around 5,000 clicks. The client’s own informational pages, on similar SERPs, earn 2% to 4% CTR in that position band, which gives 1,000 to 2,000 clicks. The commercial cluster (comparisons, pricing, “software for [use case]”) is only 8,000 searches a month, but the client’s observed CTR there is 6% to 10%, giving 480 to 800 clicks. At the client’s conversion rates (0.2% informational, 2% commercial), the informational cluster forecasts 2 to 4 leads a month and the commercial cluster 10 to 16. The commercial cluster has less than a sixth of the informational cluster’s volume but produces most of the forecast value. All figures are illustrative.

    That reallocation is often the most useful output of a forecast. It changes the plan, not just the number.

    Presenting it so it survives the first quarter

    Separate what you control from what the market controls. Pages shipped, technical fixes, rankings and query coverage are outputs you can commit to. Clicks depend on SERP layouts Google can change at any time. Agree up front that you will re-forecast quarterly, and sooner after a core update or a visible change in AI Overview prevalence in a key cluster.

    Be explicit about what you cannot measure. AI Mode clicks cannot be separated in Search Console, and practitioners report that they arrive in GA4 as google / organic. GA4’s AI Assistant channel group names ChatGPT, Gemini and Copilot as examples, but traffic without a referrer still lands in Direct. Also check the client’s market: Pew’s data is US-only, and AI feature rollouts differ by country and language, so a US-derived discount can overstate or understate the effect elsewhere.

    Watch for this:

    Do not promise that AI-referred visitors convert better to offset lower click volume. Public evidence on that is thin and mostly vendor-reported. If you want to test it, measure conversion rate by channel on the client’s own data before putting it in a forecast.

    How to tell whether the forecast or the plan is wrong

    At 90 days, compare forecast with actual clicks and conversions for each cluster. If the error sits in AI Overview-tagged clusters while rankings landed where planned, the CTR assumption was wrong: lower it and tell the client why. If rankings fell short, the plan or the competition estimate was wrong, and that is a different conversation. If Search Console impressions are growing while clicks are flat in a cluster, check the generative AI report for that page group before blaming the content, because the page may be gaining AI exposure that does not convert into visits.

    If I were pitching a new client tomorrow, I would bring a forecast built on their own post-September 2025 click rates, split by cluster and SERP type, expressed as three ranges with the assumptions written in, and weighted towards the clusters where clicks still convert. It will look smaller than a curve-based forecast. It will also still be believable in six months.

    Need help with this on your own site?

    Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.

    See SEO services

Add a reply

Replies are open to approved members. Log in or apply to join.