What changed in AI search engines this week
Week of 26 October 2026 (2026-W44). Every Monday we measure the same questions on the same engines and publish what moved, by how much and, when it is known, why.
No engine moved more than it usually does from one week to the next (3 engines, the same brands every week). In Perplexity we changed how we measure this week, so its movement is not attributed to the engine. Google AI Mode cites 4.44 sources per answer on average (n=774).
- Panel
- Up to 9 Spanish brands, always the same within each engine’s window. No brand or domain is published.
- Questions
- Buying questions in Spanish (Spain), the same every week.
- Engines
- Google AI Mode through its real interface (via a data provider); Perplexity through its API, with web search; ChatGPT through its API, without web search. One run per question and engine each week.
- Threshold
- A movement is flagged when this week’s jump is at least twice the usual week-to-week variation of the window (standard deviation of the previous jumps).
Google AI Mode
Nothing outside its usual week-to-week movement. Panel: 8 brands measured every week from 2026-W36 to 2026-W44.
| Week | Answers | Citations / answer | No citations | Domains / answer | Length | Brands named | Panel presence (95% CI) |
|---|---|---|---|---|---|---|---|
| 2026-W36 | 720 | 4.22 | 5.1% | 3.58 | 2,923 | 0.59 | 18.2 (15.5–21.2) |
| 2026-W37 | 719 | 4.06 | 6.5% | 3.45 | 2,900 | 0.62 | 19.6 (16.9–22.7) |
| 2026-W38 | 720 | 4.70 | 3.1% | 3.70 | 2,983 | 0.63 | 21.0 (18.2–24.1) |
| 2026-W39 | 720 | 4.43 | 7.4% | 3.66 | 3,076 | 0.63 | 20.1 (17.4–23.2) |
| 2026-W40 | 720 | 4.51 | 3.3% | 3.78 | 3,027 | 0.59 | 18.2 (15.5–21.2) |
| 2026-W41 | 760 | 4.48 | 4.1% | 3.77 | 3,099 | 0.61 | 20.1 (17.4–23.1) |
| 2026-W42 | 774 | 4.39 | 8.0% | 3.64 | 2,966 | 0.57 | 20.2 (17.5–23.1) |
| 2026-W43 | 773 | 4.48 | 4.7% | 3.86 | 3,185 | 0.60 | 23.7 (20.8–26.8) |
| 2026-W44 | 774 | 4.44 | 5.4% | 3.86 | 3,133 | 0.57 | 26.0 (23.0–29.2) |
Google serves shorter AI Mode answers with fewer links · 2026-08-22 → ongoing · Cause: the engine changed · no recovery.
CITATION metrics drop for every brand at once. MENTION metrics (Share of Answer, brand presence) are not affected: they are computed on the answer text, which is still complete.
Source: Confirmed by our data provider on 28 Aug 2026 over ~2,378 paired daily prompts, and reproduced by our own closed panel. No recovery seen through September.
Perplexity
Our measurement method changed this week: any movement is ours, not the engine’s. Panel: 9 brands measured every week from 2026-W36 to 2026-W44.
| Week | Answers | Citations / answer | No citations | Domains / answer | Length | Brands named | Panel presence (95% CI) |
|---|---|---|---|---|---|---|---|
| 2026-W36 | 810 | 17.04 | 0.0% | 14.10 | 1,619 | 0.65 | 15.1 (12.8–17.7) |
| 2026-W37 | 810 | 17.66 | 0.0% | 14.44 | 1,594 | 0.70 | 14.9 (12.6–17.6) |
| 2026-W38 | 810 | 18.34 | 0.0% | 15.09 | 1,619 | 0.67 | 15.9 (13.6–18.6) |
| 2026-W39 | 810 | 18.38 | 0.0% | 14.84 | 1,591 | 0.68 | 16.3 (13.9–19.0) |
| 2026-W40 | 810 | 18.52 | 0.0% | 14.03 | 1,634 | 0.61 | 15.8 (13.5–18.5) |
| 2026-W41 | 850 | 19.43 | 0.0% | 14.44 | 1,646 | 0.61 | 15.1 (12.8–17.6) |
| 2026-W42 | 864 | 9.99 | 0.0% | 8.19 | 1,742 | 0.56 | 14.8 (12.6–17.3) |
| 2026-W43 | 864 | 9.99 | 0.0% | 8.13 | 1,721 | 0.56 | 15.6 (13.4–18.2) |
| 2026-W44 | 864 | 20.27 | 0.0% | 13.84 | 1,641 | 0.50 | 14.1 (12.0–16.6) |
The Perplexity API returned at most 10 sources per answer · 2026-10-07 → 2026-10-25 · Cause: our data provider · resolved.
CITATION metrics from Perplexity (sources cited, source prominence, Authority Share, citation gap) dropped to about half for every brand at once between 7 and 25 October 2026. Share of Answer was not affected: across 12 brands and the same prompts it went from 12.6% to 13.6%, with overlapping confidence intervals. From 26 October we request 20 sources explicitly, so citation counts are not comparable across that date either.
Source: Observed in our measurement (12 brands, over 1,900 answers) and confirmed by Perplexity support on 23 Oct 2026: the old endpoint of its API lost support on 27 Sep 2026 and has since applied a default cap of 10 sources.
ChatGPT (no web search)
Nothing outside its usual week-to-week movement. Panel: 8 brands measured every week from 2026-W36 to 2026-W44.
| Week | Answers | Length | Brands named | Panel presence (95% CI) |
|---|---|---|---|---|
| 2026-W36 | 720 | 1,341 | 0.31 | 5.0 (3.6–6.8) |
| 2026-W37 | 720 | 1,350 | 0.33 | 5.1 (3.8–7.0) |
| 2026-W38 | 720 | 1,345 | 0.30 | 4.6 (3.3–6.4) |
| 2026-W39 | 720 | 1,345 | 0.31 | 5.3 (3.9–7.2) |
| 2026-W40 | 720 | 1,351 | 0.28 | 5.6 (4.1–7.5) |
| 2026-W41 | 760 | 1,356 | 0.29 | 4.7 (3.4–6.5) |
| 2026-W42 | 774 | 1,360 | 0.29 | 4.9 (3.6–6.7) |
| 2026-W43 | 773 | 1,348 | 0.32 | 4.5 (3.3–6.2) |
| 2026-W44 | 774 | 1,351 | 0.31 | 4.9 (3.6–6.7) |
ChatGPT serves its mobile-web interface and stops exposing the model and fan-out · 2026-08-25 → ongoing · Cause: the engine changed · no recovery.
NONE on our figures: we only read the answer text and its sources, and neither changes.
Source: Our data provider’s newsletter of 28 Aug 2026.
ChatGPT moves to GPT-6 and its answers can be tables, buttons or tools · 2026-10-07 → ongoing · Cause: the engine changed · no recovery.
NONE on our series: we measure ChatGPT through the API with a fixed model, which has not changed. It is logged because the gap between what we measure and what a ChatGPT user sees has grown.
Source: OpenAI official announcement of 7 Oct 2026 ("GPT-6 for everyone").
What this report does NOT prove
- It does not measure the whole market: it is a panel of Spanish brands we measure every week, mostly services and B2B, with buying questions in Spanish.
- "Outside the usual" is a descriptive threshold: a jump twice the size of the window’s weekly variation. With few reference weeks, a jump can exceed it by chance.
- It detects that something moved, not why. A cause is only stated when it is in the engine-change catalogue, with its source.
- Presence is that of the brands in this panel, not yours. Your brand can move against the panel.
Frequently asked questions
Why has my visibility in ChatGPT or Google AI Mode suddenly dropped?
There are three possible reasons and only one is yours: you lost presence, the tool that measures you broke, or the engine changed how it answers. When a whole panel of brands moves in the same week, it is almost never one brand: it is the engine.
Do Google AI Mode citations change much from one week to the next?
In our panel, Google AI Mode cited 4.44 sources per answer in 2026-W44, against 4.22 in 2026-W36. From one week to the next it usually moves by about 0.29. The large drop of August 2026 came from Google serving shorter answers with fewer links, not from the brands.
What should I do if my brand dropped and the engine did not change?
Then the change is yours, and it has to be found in your questions, your sources and your competitors. That is what a full audit separates: we measure your buying questions on four engines and tell you which part of the change is yours and which part is the engine’s.
Has your visibility dropped and you don’t know if it’s you or the engine?
This page tells you what changed in the engines. It cannot tell you what happened to your brand. That is the full audit: we measure your buying questions on four engines, separate what is yours from what is the engine’s, and hand you the plan (page in Spanish).
See the full audit →Then comes the consulting to carry out the plan, and when you want to follow it every week on your own, measurement with vilnux. Audit and consulting prices · vilnux plans
Home · Methodology · Plans & pricing · X · Product Hunt