Why Your Google Traffic Dropped — And Why Average Position Cannot Confirm the Cause
- Google Search Console reports average position as the topmost position a link to your site held, averaged across queries — and an AI Overview occupies a single position whose links all inherit it. A citation inside an AI Overview can therefore record a strong position while your ordinary result sits far lower.
- The Search generative AI performance report reports impressions for AI Overviews and AI Mode. It carries no click metric and no query dimension, so an AI-feature click-through rate cannot be derived from your own data.
- Its chart aggregates by property, and Search Console excludes Search Labs experiments — so its impression counts are not on the same basis as the main Performance report and cannot be subtracted from it.
- A dedicated opt-out toggle for generative AI features began testing on 2026-06-03 with a subset of UK site owners. Everywhere else the only documented lever is
nosnippet/max-snippet, which suppresses the ordinary web-search snippet at the same time.Google-Extendeddoes not govern AI Overviews at all.
Organic sessions on your guides are down and order volume is not. The standard reading of that pattern — rankings held, AI Overviews ate the clicks — is probably directionally right and is not something you can confirm from Search Console, because three documented properties of the reporting break the measurement. Average position does not mean what the diagnostic assumes it means. The report that covers AI features has no click metric. And the two numbers you would want to subtract are counted on different bases. This post specifies each property from Google’s own documentation, then gives the diagnostic that still works and the controls that actually exist as of August 2026.
Contents
- Your average position did not hold — it was overwritten
- How one AI Overview citation rewrites your reported position
- The generative AI report carries impressions, not clicks
- Why an AI Overview click-through rate is not derivable from your data
- The opt-out toggle shipped in June 2026 as a UK test
- Google-Extended is the wrong lever
- The lever most sites have is welded to their web-search snippet
- Google documents no special optimization for AI features
- A diagnostic runbook for a commerce blog traffic drop
- What this changes for a WooCommerce, Shopify or NetSuite team
Your average position did not hold — it was overwritten
The “rankings held, clicks fell” diagnostic rests on average position being a stable measure of where your ordinary result ranks. It is not. Google defines the metric as “the topmost position occupied by a link to your property or page in search results, averaged across all queries” (Search Console Help). Only the highest position counts; every lower appearance for the same property on that results page is discarded.
Google documents the AI Overview case explicitly in the same page: “An AI Overview occupies a single position in search results, and all links in the AI Overview are assigned that same position.” Put the two rules together and the consequence is mechanical. If an AI Overview cites your page and your ordinary blue link also ranks, Search Console records the AI Overview’s position and silently drops the blue link’s. A page whose organic result slid from position 4 to position 9 can report an improved average position for the same period, purely because it started being cited in an AI Overview near the top of the page.
That is the reverse of the story the metric is usually asked to tell. Stable or improving average position alongside falling clicks is treated across the search-marketing corpus as proof that ranking is intact and AI answers are absorbing the clicks. It is equally consistent with the ordinary ranking having fallen and the measurement hiding it.
How one AI Overview citation rewrites your reported position
The masking effect needs no unusual conditions — a single citation is enough. The diagram below traces one query through both rules.
The practical loss is attribution. A drop caused by a genuine ranking decline and a drop caused by an AI Overview absorbing the click produce the same shape in the report — falling clicks against flat or improved average position — and call for opposite responses. One is a content and relevance problem. The other is a presentation change you do not control. Both sit inside the wider shift covered in the AI for commerce teams guide hub.
The generative AI report carries impressions, not clicks
Search Console does now report AI-feature visibility separately. The Search generative AI performance report covers, in Google’s words, “impressions for the following generative AI capabilities on Google Search: AI Overviews, AI Mode” (Search Console Help). Impressions are defined there as “how many times links to your site were shown to a user in a generative AI feature on Google Search.”
What the report does not carry is the metric the traffic question turns on. There is no click metric, and there is no query dimension. The available dimensions are pages, countries, devices and dates. So the report answers “how often was my site shown inside an AI feature” and cannot answer “how often did anyone click through from one”, or “which queries did it happen on”.
Availability is also gated twice. Google states it is rolling the report out “to a subset of website owners, allowing for thorough testing before rolling it further”, and a property must have received enough impressions in generative AI features for the report to appear at all. A small commerce blog — the category most affected by the traffic shift — is the category least likely to clear that threshold.
Why an AI Overview click-through rate is not derivable from your data
The obvious workaround is arithmetic: take total clicks from the main Performance report, take AI-feature impressions from the generative AI report, and derive the rest. Two documented counting rules make that invalid.
The first is aggregation. Google states that in the generative AI report, “if two results from the same site appeared in a generative AI search results feature, they count as a single impression in the chart total” — the chart aggregates by property. The main Performance report exposes both property-level and page-level aggregation, and its totals are built on a different rule for the same event. Subtracting one report’s number from the other’s compares two different countings of the same page view.
The second is scope. Google states plainly that “Search Console doesn’t include data from experiments in Search Labs, as these experiments are still in active development.” Exposure inside an experimental AI surface is real exposure that produces no row anywhere in your reporting. The size of that gap is not published, which means it cannot be corrected for.
The honest position: the widely-quoted click-through-rate figures for AI Overviews come from third-party panel studies, not from first-party data any site owner can reproduce. Treat them as directional heuristics from vendor studies, revisit them quarterly, and do not attempt to reconcile them against your own Search Console numbers — the reconciliation is not defined.
The opt-out toggle shipped in June 2026 as a UK test
Google announced a dedicated control on 2026-06-03. The company described “a new control that lets website owners manage how their links and content appear in generative AI Search features”, surfaced in Search Console (Google blog). Two properties of it matter more than its existence.
The first is the cost, stated without hedging: “Sites that opt out will not receive traffic or impressions from our generative AI features.” This is an exclusion, not a preference — AI Mode and AI Overviews stop being a channel entirely. The second is the decoupling, which is the genuinely new part: “This control will not be used as a ranking signal for search results outside of these generative AI Search features.” Ordinary organic ranking is explicitly untouched.
The caveat is the reason most readers cannot act on this. Google stated it is “beginning to roll these features out to a subset of website owners in the UK, allowing for thorough testing before rolling them out to website owners globally”. As of August 2026 that is a geographically scoped test, not a general control. Check for the toggle in Search Console before planning around it, and do not treat secondary coverage describing it as generally available as accurate for your property.
Google-Extended is the wrong lever
The most common mistaken response is to disallow Google-Extended in robots.txt and consider AI Overviews handled. Google documents that this control governs a different set of products: training and grounding for “Gemini Apps and Grounding with Google Search on Vertex AI” (Google crawler documentation). AI Overviews and AI Mode are not in that set.
Two further details are worth holding. Google states that “Google-Extended does not impact a site’s inclusion in Google Search nor is it used as a ranking signal in Google Search” — so disallowing it is safe for ranking, and equally ineffective against the feature people deploy it against. And Google-Extended has no separate HTTP user-agent string; it is a robots.txt token applied to crawling that Google’s existing crawlers already perform.
Google’s own AI-features guidance points at Googlebot instead: “AI is built into Search and integral to how Search functions, which is why robots.txt directives for Googlebot is the control for site owners to manage access to how their sites are crawled for Search” (AI features and your website). Disallowing Googlebot removes the site from Search altogether. There is no partial setting at the crawler layer.
The lever most sites have is welded to their web-search snippet
Outside the UK test, the documented controls that reach AI features are the ordinary preview controls — and they are not scoped to AI features. Google’s robots meta tag reference states that nosnippet “applies to all forms of search results (at Google: web search, Google Images, Discover, AI Overviews, AI Mode)” and “will also prevent the content from being used as a direct input for AI Overviews and AI Mode” (robots meta tag reference).
That single sentence is the whole trade. Using nosnippet to keep a page out of AI Overviews also removes its text snippet from ordinary web search, Images and Discover. max-snippet is not an escape either: Google documents the special value “0: No snippet is to be shown. Equivalent to nosnippet.” A snippet-length limit above zero simply caps the text available everywhere, AI features included.
| Control | Reaches AI Overviews / AI Mode | Cost elsewhere | Available to you now |
|---|---|---|---|
Google-Extended |
No — governs Gemini Apps and Vertex AI grounding | None; not a ranking signal | Yes |
| Disallow Googlebot | Yes — by leaving Search | Total: removes the site from Search | Yes |
nosnippet / max-snippet:0 |
Yes — blocks use as a direct input | Removes the text snippet in web search, Images, Discover | Yes |
| Search Console generative AI toggle | Yes — no traffic or impressions from AI features | None stated for ordinary ranking | Subset of UK owners, in testing |
Verdict: for a commerce blog whose guides earn ordinary snippets, none of the generally available controls is worth pulling. The only one that does what operators actually want — out of AI features, intact in Search — is the toggle, and it is a scoped test. Do nothing at the controls layer and re-check availability rather than trading a live snippet for a speculative recovery.
Google documents no special optimization for AI features
The advice market around this traffic shift sells AI-specific optimization as a discipline. Google’s guidance for AI features, last updated 2025-12-10, states the opposite: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.”
Read that precisely, because it is narrower than “nothing changed”. It says eligibility carries no separate technical checklist — the same crawlability, indexability and preview settings govern both surfaces. It does not say the traffic outcome is unchanged, and it does not say content strategy is unaffected. What it removes is the premise that a distinct markup or configuration layer is required to be eligible.
The strategic response therefore sits in content and measurement, not configuration. Which content earns citation, and how a commerce team should read its own numbers, is covered in the companion piece on how answer engines select the sources they cite. The configuration layer, per Google’s own page, needs nothing done to it.
A diagnostic runbook for a commerce blog traffic drop
Work down this list in order. It is built to separate the causes that look identical in the Performance report, and every step reads from data a property owner already has.
- Confirm the drop is in clicks, not impressions, and record both figures for the same date range before changing anything.
- Split the affected URLs by intent: informational guides versus product, category and transactional pages. A drop concentrated in guides is consistent with AI-feature absorption; a drop across transactional pages is not, and points elsewhere.
- Stop reading average position as a ranking measure on any URL that may be cited in an AI Overview — the topmost rule makes it ambiguous.
- Re-measure ranking outside Search Console for a sample of affected queries, so blue-link position is observed rather than inferred.
- Open the Search generative AI performance report if the property has it, and read impressions by page only — no click or query data exists there to read.
- Do not subtract generative AI impressions from Performance report totals; the chart aggregates by property and the two figures are not on the same basis.
- Check
robots.txtfor aGoogle-Extendeddisallow added as an AI-Overviews countermeasure, and confirm what it was intended to achieve — it does not affect AI Overviews. - Audit templates for a
nosnippetor lowmax-snippetvalue applied site-wide, which suppresses ordinary snippets as well as AI-feature use. - Verify revenue and order volume over the same window before treating the drop as a commercial problem rather than a channel-mix change.
- Re-check Search Console for the generative AI opt-out toggle on a quarterly cadence rather than acting on secondary reports of its availability.
What this changes for a WooCommerce, Shopify or NetSuite team
For a team running a store on WooCommerce or Shopify with NetSuite behind it, the operational change is narrow and specific: the content-performance number that feeds planning is no longer a single readable series, so stop planning against average position and start planning against the two things that remain unambiguous — clicks by URL, and orders.
Concretely, that means three adjustments. Segment reporting by page intent so a guide-traffic decline never gets read as a storefront problem. Treat any average-position series on a guide URL as ambiguous rather than as evidence, and use it to trigger an out-of-console ranking check instead of a content rewrite. Keep an explicit note of which measurement gaps are structural — no click metric for AI features, no Search Labs data, no query dimension — so the gap does not get filled with an assumption at the next planning cycle.
The commercial pattern underneath is the one worth acting on. If informational sessions fall while order volume holds, the channel has shifted rather than shrunk, and the correct response is to protect the transactional and product surfaces that still convert, not to rewrite guides against a metric that cannot confirm the diagnosis. Where AI systems are already reaching the store directly, the buyer-facing mechanics are covered in the analysis of what autonomous shopping agents mean for a store. The longer arc for integration teams is set out in the 2027 integration outlook.
Get the working checklists
The runbooks and decision checklists from these guides, as printable PDFs — free in the SoftXone guide library.
References
- Search Console Help — What are impressions, position, and clicks?Google — defines average position as the topmost position averaged across queries, and states that an AI Overview occupies a single position whose links all share it.
- Search Console Help — Generative AI performance report (Search)Google — states the report covers impressions for AI Overviews and AI Mode, lists its four dimensions, describes property-level chart aggregation, and excludes Search Labs data.
- Google — New opportunities, control and insights for website ownersGoogle blog, 3 June 2026 — announces the generative AI opt-out control, its effect on traffic and impressions, and the UK-subset rollout.
- Google Search Central — Robots meta tag specificationGoogle — defines nosnippet and max-snippet, their application to AI Overviews and AI Mode, and the equivalence of max-snippet:0 to nosnippet.
- Google Search Central — Google common crawlersGoogle — documents Google-Extended as a control for Gemini Apps and Vertex AI grounding, and states it does not affect Search inclusion or ranking.
- Google Search Central — AI features and your websiteGoogle — states that robots.txt directives for Googlebot are the control for Search AI features, and that no additional requirements or special optimizations exist.
Frequently asked questions
Does a stable average position mean my rankings held?
Not necessarily. Search Console reports the topmost position a link to your property held, averaged across queries, and an AI Overview occupies a single position that every link inside it shares. If your page is cited in an AI Overview near the top of the results while its ordinary result ranks lower, the report reflects the AI Overview position and discards the lower one. Confirm blue-link position outside Search Console before concluding that ranking was unaffected.
Can I see how many clicks AI Overviews sent to my site?
No. The Search generative AI performance report covers impressions for AI Overviews and AI Mode, with dimensions for pages, countries, devices and dates. It carries no click metric and no query dimension. Clicks arriving from an AI feature are not broken out anywhere in Search Console, so any AI-feature click-through rate quoted in the trade press comes from third-party panel studies rather than from data a site owner can reproduce.
Will blocking Google-Extended remove my site from AI Overviews?
No. Google documents Google-Extended as a control for training and grounding in Gemini Apps and Vertex AI, and states that it does not affect a site inclusion or ranking in Google Search. AI Overviews and AI Mode are Search surfaces, governed by the robots.txt directives that apply to Googlebot. Disallowing Google-Extended is safe for ranking and does not change AI Overview eligibility.
Is there a way out of AI Overviews without losing my search snippet?
Only through the Search Console control Google began testing on 3 June 2026, which Google said would not be used as a ranking signal for results outside generative AI features. That test began with a subset of UK site owners. Everywhere else the documented lever is nosnippet, or max-snippet set to zero, which Google defines as equivalent and which also suppresses the text snippet in ordinary web search, Images and Discover.
Should I rewrite my guides in response to the traffic drop?
Not on the strength of a Search Console reading alone, because the report cannot separate a genuine ranking decline from AI-feature absorption. Segment the affected URLs by intent first, confirm whether order volume moved alongside sessions, and check blue-link position outside the console. A channel-mix change and a relevance problem produce the same shape in the report and call for different work.
Leave a Reply