AI search is becoming an operating surface for brand teams. The practical question is not whether a single model response changed—it is whether the change reveals something your team can verify and improve.
What changed
On September 16, SISTRIX reported that further changes to the way Google delivers search results had reduced its data-collection rate. Two days later, Search Engine Roundtable reported that Nozzle’s Derek Perkins had seen about an 80% drop in the Google data his tools could retrieve, while SISTRIX was still adapting its collection process. The reports describe a change in the ability of third-party tools to obtain and process result data. They do not establish that the underlying ranking, an AI Overview, an AI Mode answer, or a site’s traffic changed by the same amount.
This is a monitoring story, not a confirmed search-ranking update
Google has not published a corresponding announcement that explains the September delivery changes or labels them a ranking update. SISTRIX says its data is still flowing and its Toolbox is updating, albeit at a reduced rate. Search Engine Roundtable’s account is independent reporting based on vendor observations, not a platform confirmation. That distinction matters: an abrupt break or delay in a rank-tracking chart can be caused by the measurement path before it is evidence of changed visibility.
The context is a longer shift in how results are collected
SISTRIX’s status history describes a series of changes that have made automated result collection more difficult over roughly the past year. It previously documented Google’s removal of the num=100 parameter and, more recently, result links routed through Google redirect URLs. Those changes can add processing steps or requests for tools that need to identify target domains and positions. The current notice does not disclose the precise mechanism behind the September 16 slowdown, so it would be premature to attribute it to one URL format, a particular anti-bot system, or an AI-search feature.
Why an 80% retrieval figure is easy to misread
The reported 80% number is Nozzle’s observation of data it could retrieve, as relayed by Search Engine Roundtable. It is not a measurement of an 80% loss in organic clicks, impressions, AI citations, or brand recommendations. It also is not a cross-tool benchmark: SISTRIX described a reduced rate, not the same percentage, and the report says effects may vary across providers. Treat the figure as evidence that collection conditions may be materially disrupted for at least one tool—not as a reusable business metric.
AI-search teams need to separate the surface from the instrument
AI visibility work already combines several imperfect records: an answer capture, displayed sources, referral data, Search Console, merchant or ad-platform data, and third-party monitoring. A disruption in one record should not overwrite the others. If a tracked Google or AI-search series moves abruptly this week, preserve the date, market, device, query set, provider, and collection status. Then compare it with first-party performance data and direct captures of the user-facing result before calling the movement a platform change.
What first-party data can and cannot settle
Google Search Console can help test whether clicks and impressions moved alongside a third-party chart. Where available, its generative-AI performance reporting can add a Google-owned visibility view for AI features. But first-party data still does not reveal every answer text, citation, recommendation, competitor, or collection condition. Conversely, a clean traffic series does not prove that every tracked position is unchanged. The practical response is triangulation: use each source for the part of the journey it can actually observe.
A sensible operating check for this week
Flag abrupt changes rather than automatically adding them to a trendline. Ask the vendor whether data is delayed, sampled, backfilled, or recalculated; check the same query family in another collection source where the methodology is understood; review Search Console and analytics for the matching market and time window; and capture a small, controlled set of live results for high-value prompts. Record the outcome as confirmed, unsupported, or still unknown. That light process keeps a reporting anomaly from turning into a rushed content, media, or executive decision.
What remains unknown
Public evidence does not yet show exactly what Google changed in mid-September, how widely it affects providers, how long adaptation will take, or whether any historical measurements will be revised. It also does not show a direct effect on AI Overviews, AI Mode, or a brand’s visibility in those experiences. Until vendors publish more detail or Google comments, claims about the cause, scope, or commercial impact should remain provisional.
Bottom line
The current signal is about observability. SISTRIX has confirmed a reduced collection rate, and independent industry reporting has documented a sharp retrieval decline at another provider. That is enough to add caution to weekly visibility reporting; it is not enough to declare a Google ranking or AI-search shift. Keep the raw evidence, compare independent records, and let repeated user-facing observations—not a single disrupted chart—drive the next action.