AI Search Ranking: Volatility, Stability and What the Data Actually Shows

Introduction: Twinings Gained 39.6 Percentage Points in Two Weeks. Should Anyone Care?

Between December 20 and January 3, Twinings gained 39.6 percentage points on ChatGPT. Its mention rate went from mid-pack to dominant in the space of two weeks. If that were a stock movement, it would trigger alerts. If it were a Google ranking shift, it would prompt emergency calls to the SEO team. In traditional search, a movement that large in two weeks represents a meaningful event — an algorithm update, a manual action, a significant new backlink profile.

But in AI search ranking, it may be noise.

The same brand — Twinings — also held the best average position on ChatGPT at all five measurement points across a ten-week longitudinal study. Through all the volatility, through all the swings in mention rate, through all the platform fluctuations that moved other brands up and down, Twinings’ positional prominence remained consistent. That is signal. The 39.6-point mention rate swing that happened in two weeks? That is what the data suggests was probably noise.

Luther and Touboul-Cohen (2026) conducted the first longitudinal study of AI visibility metrics for real competing brands in a real product category — six major U.S. tea brands, two AI platforms, 50,000+ individual AI responses over ten weeks. The volatility findings they document are the most practically consequential in the study, because they directly determine how businesses should respond to the AI search ranking data they collect. If every fluctuation is signal, constant tactical adjustment is rational. If most fluctuations are noise concealing a durable underlying structure, constant adjustment is wasteful and potentially counterproductive.

This post maps the volatility data, explains the stability that persists beneath it, and provides the signal-versus-noise framework that AI search ranking monitoring requires.

Quick Answer AI search ranking is highly volatile — individual brands experience mention rate swings exceeding 30 percentage points in two weeks, and Google AI Overviews shows 50% more volatility than ChatGPT. But beneath the volatility, a durable competitive hierarchy persists, confirmed by Kendall’s W concordance values of 0.785 (ChatGPT) and 0.743 (Google AI Overviews). Most single-interval fluctuations are noise. Directional patterns across three or more intervals are signal.

What Is AI Search Ranking and How Is It Measured?

AI search ranking is a fundamentally different construct from traditional search ranking, and the measurement challenge is correspondingly different.

In traditional SEO, rank position is a discrete, deterministic measurement: a brand either occupies position 1, 2, or 3 for a given keyword, and that position is the same for any user running the same query in the same context. Rank tracking tools query Google or Bing directly and return a precise position number. A rank tracking report from Monday tells you something reliable about where you will rank on Tuesday, absent an algorithm update.

AI search ranking operates differently across two dimensions. First, it is non-deterministic: the same query run in fifty independent sessions on the same day produces fifty potentially different responses. A brand might appear in 43 of those sessions and not appear in the other 7. Its “rank” is not a single number — it is a distribution across sessions, expressed as a mention rate (how often it appears) and an average position (where it appears when it does).

Second, it is generative rather than retrievive. Traditional search engines retrieve pre-existing rankings. AI systems generate responses through probabilistic processes that draw on learned associations, training data, retrieval context, and response-generation logic. The same brand query processed through the same model on the same day can produce meaningfully different outputs across sessions depending on random seed values, retrieval context, and the specific phrasing of the prompt.

This architecture makes AI search ranking inherently variable in a way that traditional search is not — and makes single-session, single-interval measurements unreliable as indicators of actual platform performance.

Luther and Touboul-Cohen (2026) addressed this measurement challenge by running 50 independent sessions per prompt per measurement interval, producing statistically meaningful mention rate estimates rather than single-session snapshots. Across 120 data points — six brands, two platforms, two metrics, five dates — over 50,000 individual AI responses were collected. The volatility documented in this study is not measurement error; it is the genuine behavior of AI search ranking systems across real brands over real time.

For the broader context of how AI search differs from traditional search as a visibility construct, see AI search. For how generative engine optimization functions as a strategic discipline within AI search environments, the foundational framing applies directly.

How Volatile Is AI Search Ranking?

The volatility data from Luther and Touboul-Cohen (2026) establishes the empirical baseline for what “volatile” actually means in AI search ranking contexts.

The mean coefficient of variation for mention rate across the study was 22.2% on ChatGPT and 33.9% on Google AI Overviews. Coefficient of variation measures variability relative to the mean — a CoV of 22.2% means the typical brand’s mention rate varies by about a fifth of its mean value across measurement intervals. A brand with a mean mention rate of 40% might range from 31% to 49% across the measurement period. On Google AI Overviews, the same brand might range from 27% to 53%.

At the individual brand level, the volatility is more dramatic. Single-interval swings exceeding 30 percentage points were documented — the most notable being Twinings on ChatGPT gaining 39.6 percentage points between December 20 and January 3. Google AI Overviews shows approximately 50% more volatility than ChatGPT on both the mention rate and average position metrics.

To contextualise this: in traditional search ranking, a movement of 5–10 positions for a competitive keyword within a two-week period would be considered highly significant and would typically prompt investigation. A movement of 30–40 percentage points in AI mention rate within the same timeframe is, according to the longitudinal pattern data, entirely consistent with normal AI search ranking behavior. This is the fundamental reason why AI search monitoring methodology must differ from traditional rank tracking methodology.

For the analysis of how AI search platforms differ structurally — including why Google AI Overviews shows more volatility than ChatGPT — see AI search platforms.

What Persists Beneath the Volatility?

The most practically important finding in the volatility analysis is not the volatility itself — it is what survives beneath it. Luther and Touboul-Cohen (2026) computed Kendall’s W rank concordance across the five measurement dates: the concordance value for mention rate is 0.785 on ChatGPT and 0.743 on Google AI Overviews. These are moderate to strong concordance values, indicating that the relative ordering of brands — which brand ranks first, second, third — remained substantially consistent across the five measurement intervals even as the specific numerical values fluctuated substantially.

This is the key insight for AI search ranking strategy: the surface volatility is real, but it sits on top of a more stable underlying competitive structure. The brands that lead on average lead consistently. The brands in the middle of the pack stay there on average. The competitive hierarchy shifts at the margins but does not reverse chaotically.

The most direct evidence of this underlying stability is Twinings’ average position performance. Despite all the mention rate volatility in the dataset — including Twinings’ own 39.6-point swing — Twinings held the best average position on ChatGPT at every single one of the five measurement intervals. This is the only instance of sustained single-metric leadership in the entire dataset. Whatever produced Twinings’ positional prominence on ChatGPT, it was consistent enough to survive ten weeks of AI system behavior, model updates, and competitive movement. That consistency points to genuine underlying signal — content quality, brand authority, and earned media presence that AI systems consistently evaluate as high-confidence — rather than algorithmic luck.

The stock market analogy from Luther and Touboul-Cohen is precisely right: “A brand that sees its mention rate fall 15 percentage points between two intervals has not necessarily lost long-term ground if the broader pattern across several measurements remains stable or upward. Volatility findings do not require constant tactical adjustment but do warrant sustained observation for emerging directional patterns.”

This connects to the durable signal framework from Kargaev (2026): Brand Entity Mentions scoring NIS 0.918 as the dominant GEO signal, topical authority and factual accuracy showing Very Strong cross-paradigm correlations in Iyappan (2026). These foundational signals do not produce dramatic week-to-week fluctuations. They produce the kind of consistent underlying authority that keeps a brand in the leading position across five consecutive measurement intervals despite surface volatility.

For the foundational signals that drive stable AI search ranking performance, see brand entity SEO. For the full SEO vs GEO analysis, see SEO vs GEO.

Why Does AI Search Ranking Behave Differently From Traditional Search Ranking?

The volatility of AI search ranking is not a bug or an immature feature that will disappear as AI platforms mature. It is a structural consequence of how generative AI systems work.

Traditional search ranking is deterministic. Google’s algorithm applies a fixed set of rules and signals to rank pages. The same page, evaluated with the same signals, in the same context, produces the same rank. This determinism makes traditional SEO ranking relatively stable and makes rank tracking straightforward: track the same keywords over time in the same conditions and you observe the same underlying reality.

AI search ranking is probabilistic and generative. When ChatGPT generates a response to a tea recommendation query, it samples from probability distributions shaped by its training data, its learned associations, and its response-generation parameters. The output is a text string that is statistically likely given the input — not a deterministic retrieval of a pre-established ranking. Different random seeds, different sampling temperatures, different retrieval contexts produce different outputs. This structural non-determinism is what produces the measured volatility — and it cannot be eliminated by any optimization strategy.

Kargaev (2026) provides the organic foundation framing: AI systems draw from the indexed, organically-visible web, so SEO foundations matter. But the generative layer adds stochastic variation that traditional ranking never had. Iyappan (2026) documents the performance differences across eight metrics between SEO-calibrated content (low AI retrieval compatibility) and GEO-calibrated content (high AI retrieval compatibility) — but even GEO-calibrated content does not produce deterministic AI search rankings because the underlying generation process is not deterministic.

The practical implication is that AI search ranking strategy must be built around influencing statistical tendencies rather than controlling specific outcomes. A brand cannot guarantee that it will appear in position one for every tea recommendation query. It can build the content quality, brand entity signals, and earned media presence that statistically increases the probability of prominent, confident AI citations across the range of relevant queries. That is what Twinings appears to have done — and what the Kendall’s W concordance data confirms as the operating logic of durable AI search ranking performance.

For the SEO AEO GEO paradigm comparison that explains how each optimization stage contributes to AI search ranking performance, see SEO AEO GEO.

What Causes AI Search Ranking Volatility?

Understanding what produces AI search ranking volatility is useful for distinguishing genuine signal from noise — because not all volatility has the same source.

Model updates. LLM providers push model updates that can shift learned associations and response generation behavior. A ChatGPT model update that changes how the model weights brand familiarity versus product-specific knowledge can produce mention rate shifts that have nothing to do with anything the brand has done. These updates are the most common source of genuine directional signal — a persistent shift following a known model update warrants strategic attention.

Training data recency and weighting. As new content is incorporated into model training data, the associations between brand names and topic areas shift. A brand that receives substantial positive editorial coverage in a short period may see mention rate improvements as that coverage enters the training data distribution. Conversely, a brand that stops generating new editorial content may see gradual decline as its training data becomes proportionally less represented.

Query context sensitivity. AI responses vary based on the exact phrasing of queries, the session context, and the specific framing of the recommendation request. The Luther and Touboul-Cohen methodology carefully controlled for query design — but in organic consumer behavior, subtle variations in how buyers phrase queries produce genuine variation in which brands AI systems include. This is irreducible noise.

Platform algorithm changes. Google AI Overviews is subject to policy and algorithm changes that affect which content and brands are sampled and how responses are structured. These can produce sudden shifts that persist — genuine signal — rather than the organic volatility that fluctuates around a stable mean.

Competitive content ecosystem changes. When a competing brand publishes significant new content, earns prominent editorial mentions, or launches a digital PR campaign, the competitive content ecosystem shifts. AI systems that draw on that ecosystem reflect the shift over time. This is the mechanism through which GEO investment by competitors produces mention rate changes for other brands — and why AI search ranking is a genuinely competitive, dynamic system.

For the AI optimization strategy framework that addresses how to build the signals that drive durable ranking performance, see AI optimization strategy.

How Should Businesses Respond to AI Search Ranking Volatility?

The volatility data from Luther and Touboul-Cohen (2026) supports a specific strategic response framework — one that is different from how most businesses currently manage SEO performance signals.

Rule 1: Single-interval changes are noise until confirmed. A 15-point drop in mention rate between two monthly measurements is not a strategic emergency. It is a data point. Respond to it by increasing monitoring frequency for the next interval — confirming whether the drop persists, reverses, or deepens. A drop that persists across two consecutive measurements is beginning to look like signal. A drop that reverses is noise.

Rule 2: Directional patterns across three or more intervals are signal. The most reliable AI search ranking signals are those that persist across multiple measurement intervals in the same direction. A brand whose mention rate has declined at three consecutive monthly measurements — from 45% to 38% to 31% — is showing a directional pattern that warrants strategic response. The Kendall’s W data confirms that the underlying competitive hierarchy is relatively stable, meaning persistent directional moves typically reflect genuine content ecosystem changes rather than random variance.

Rule 3: Platform-specific analysis is essential. A decline on ChatGPT and a simultaneous gain on Google AI Overviews are not the same signal and should not be averaged into a combined metric. Each platform has a different volatility profile — Google AI Overviews shows 50% more volatility than ChatGPT — and different causes for its movements. Platform-specific analysis is the prerequisite for distinguishing meaningful shifts from cross-platform averaging that obscures both.

Rule 4: Build durable signals, not tactical responses. The brands that demonstrate stable underlying position despite surface volatility — the Twinings-pattern — are building the foundational signals that AI systems consistently evaluate positively: content quality, brand entity verification, topical authority depth, high-authority earned media. These are not tactical responses to weekly fluctuations. They are sustained investment programmes that produce statistical tendencies toward prominent, confident AI citation.

For the GEO checklist that covers the foundational signal-building programme, see GEO checklist. The Google AI optimization guide provides the Google-specific framing for what content signals Google AI Overviews evaluates.

What Does the Twinings Pattern Tell Us About Durable AI Search Ranking?

The Twinings average position finding is the most practically instructive data point in the entire Luther and Touboul-Cohen study — not because it is the most dramatic but because it is the only instance of sustained, consistent performance that persists through all the volatility.

Across all five measurement intervals, across ten weeks, through all the surface fluctuations including Twinings’ own 39.6-point mention rate swing, the brand consistently occupied the best average position on ChatGPT. Every other brand in the study showed more variable positional performance. Twinings alone held its position consistently.

What the data cannot establish is exactly why. This is an observational study — it documents the pattern but cannot identify the causal mechanism with certainty. What can be said is that whatever drove Twinings’ consistent positional prominence was stable enough to survive two and a half months of AI system behavior, model updates, and competitive activity. Stability of that duration in a non-deterministic system points to the same foundational signals that Kargaev (2026) and Iyappan (2026) identify as the highest-confidence AI visibility investments: brand entity depth, factual accuracy, topical authority, and the kind of consistent earned media presence that builds the cross-referenced, verified brand identity that AI systems cite with confidence.

The Twinings pattern is not an instruction to copy Twinings’ specific strategy. It is a data point confirming that durable AI search ranking is achievable — that the underlying competitive hierarchy is not purely random — and that the path to it runs through foundational brand authority rather than algorithmic optimization.

For the full AI search content strategy that maps how foundational content investments build durable AI search ranking, the research-backed framework covers the full implementation. The ChatGPT interface provides the starting point for manual prompt testing to begin building your AI search ranking baseline.

What Does AI Search Ranking Volatility Mean for Content Investment Decisions?

The volatility finding has a direct implication for how content investments should be evaluated — one that most businesses have not yet worked through.

In traditional SEO, content investments are evaluated against ranking movements: did publishing this guide improve our position for target keywords? The evaluation cadence matches the stability of traditional rankings — changes are visible within weeks and persist long enough to be confidently attributed to specific content actions.

In AI search ranking, the same evaluation logic breaks down. A significant content investment — publishing a comprehensive topical authority guide, launching a digital PR programme, completing a brand entity optimisation — may produce genuine AI search ranking improvements that are obscured by the natural volatility of the measurement. A business that publishes a major guide in October, measures AI ranking in November, and finds no improvement may conclude the investment failed. The actual effect may be present but masked by the coefficient of variation that the data shows is 22% on ChatGPT and 34% on Google AI Overviews.

The correct evaluation framework for AI search ranking investments is slower and more patient than traditional SEO evaluation. Investments should be assessed across a minimum three-to-six month window, using the directional trend rather than the specific November-to-December delta as the evaluation metric. A brand whose mention rate trend is directionally upward across six months — even if with significant single-interval fluctuations — is showing that its content investments are working.

This patience requirement has a practical implication for how AI search ranking monitoring is reported internally. Monthly reports that show single-interval changes will generate reactive discussions about why the number went up or down. Quarterly reports that show three-month rolling trends — identifying directional patterns rather than point-to-point changes — generate strategic discussions about whether the investment programme is working. The data from Luther and Touboul-Cohen (2026) suggests that quarterly rolling trend reporting is the right cadence for executive-level AI search ranking review, with monthly monitoring maintained at the operational level for early detection of genuine directional shifts.

For the complete AI optimization strategy that addresses investment evaluation across all four paradigm stages, see AI optimization strategy.

How Does AI Search Ranking Stability Connect to Brand Reputation?

The most consequential insight in Luther and Touboul-Cohen’s (2026) study may be the finding they frame as the study’s ultimate conclusion: “AI visibility is a metric worth tracking precisely because it reflects what consumers encounter when they turn to AI systems to learn about a brand, a category, or a purchase decision. But the underlying requirement is not algorithmic. Brands that produce genuinely expert, accurate, and useful content, and that do so consistently enough to earn coverage from sources AI platforms treat as authoritative, will find that AI visibility follows.”

This conclusion connects AI search ranking stability to something older and more fundamental than any optimization technique: brand reputation as accumulated through consistent, genuine content quality and earned editorial coverage. The durable competitive hierarchy that Kendall’s W confirms — the Twinings pattern that persists through ten weeks of volatility — is not the product of GEO techniques applied since 2024. It is more plausibly the product of brand authority built over much longer periods, recognised and surfaced by AI systems that draw on the full accumulated content ecosystem.

Aggarwal et al. (2024) established that citation addition and statistical enrichment produce GEO visibility improvements in controlled settings. Kargaev (2026) quantified brand entity mentions as the dominant GEO signal at NIS 0.918. Iyappan (2026) confirmed topical authority and factual accuracy as the Very Strong cross-paradigm signals. What Luther and Touboul-Cohen (2026) add is the longitudinal confirmation that these signals produce not just higher visibility but durable visibility — the kind that holds under volatility.

For businesses building AI search ranking strategy, this is both a realistic commitment and a commercial opportunity. The realistic commitment is that durable AI search ranking requires genuine brand authority investment — sustained content quality, consistent earned media, verified entity signals — not short-term tactical interventions. The commercial opportunity is that the businesses willing to make that commitment are building a competitive position that is genuinely harder to displace than any position achieved through algorithmic tactics alone.

For the brand entity SEO signals that anchor durable AI search ranking, the research-backed framework covers the complete implementation programme.

What Does the Commercial Data Show About Why AI Search Ranking Matters Now?

The volatility finding in isolation might suggest that AI search ranking is too unstable to invest in seriously. The commercial growth data makes the opposite case.

Adobe Analytics (2025) documented a 3,500% increase in U.S. retail site traffic from generative AI sources between July 2024 and May 2025. The brands whose AI search ranking was building during that period were receiving a share of that traffic. The brands absent from AI responses were not. The stability beneath the volatility — the durable competitive hierarchy that Kendall’s W confirms — means that the competitive positions being built now are not temporary. They compound.

Pew Research Center (2025) found that users encountering AI-generated summaries clicked traditional search results only 8% of the time versus 15% for users who did not encounter an AI summary. When a buyer consults ChatGPT or Google AI Overviews for a recommendation and receives a direct, synthesised answer, the probability that they then click through to evaluate each brand separately drops by nearly half. The brands in the AI-generated answer have captured buyer attention at the highest-intent moment. The brands absent from it are competing for the 8%.

These commercial figures make AI search ranking less a future investment and more a present competitive necessity. The volatility data does not change this assessment — it refines it. The brands building durable AI search ranking through genuine content quality and brand authority are not chasing unstable metrics. They are building the consistent underlying authority that the longitudinal data confirms is achievable and commercially valuable.

How Does AIO Clicks Help Navigate AI Search Ranking Volatility?

AIO Clicks is a premium digital visibility agency headquartered in Haaksbergen, Netherlands, serving businesses across the EU. The AI search ranking volatility finding from Luther and Touboul-Cohen (2026) directly informs how AIO Clicks structures client monitoring and reporting.

Every AI search ranking fluctuation a client observes is evaluated against the signal-versus-noise framework: is this a single-interval anomaly, a platform-specific shift, or a directional pattern across multiple intervals? The answer determines whether the response is monitoring adjustment (watch for confirmation) or strategic adjustment (investigate and address the underlying driver). Managing this distinction is what separates effective AI visibility monitoring from constant tactical reaction to non-deterministic noise.

The foundational signal investments — brand entity, topical authority, content quality, digital PR — are built to produce the Twinings pattern: consistent underlying position that persists through surface volatility. Monitoring confirms the investments are working. It does not drive them.

AIO Clicks Services

AI Search & GEO — systematic AI search ranking monitoring across ChatGPT, Google AI Overviews, and Perplexity, combined with the foundational signal work that builds durable ranking performance.

Google Rankings & SEO — the organic foundation that feeds AI search ranking performance. Kargaev’s (2026) organic foundation effect confirms that SEO foundations are the prerequisite for AI retrieval eligibility.

Run the free analysis to establish your current AI search ranking baseline — mention rate and average position across platforms, results in 60 seconds.

Frequently Asked Questions About AI Search Ranking

What Is the Key Takeaway on AI Search Ranking?

The volatility of AI search ranking is real, documented, and structurally inherent to how generative AI systems work. It cannot be optimised away, and it should not be managed by constant tactical responses to week-to-week fluctuations. The data shows that most single-interval movements are noise.

What survives the volatility is the underlying competitive hierarchy — the Kendall’s W concordance values of 0.785 and 0.743, the Twinings pattern of consistent positional prominence across all five measurement intervals, the brands that have built genuine content quality and brand authority finding their position confirmed repeatedly even as specific values fluctuate.

AI search ranking strategy is not about chasing the algorithmic signal in this week’s mention rate data. It is about building the foundational authority — brand entity depth, topical expertise, factual accuracy, earned media presence — that statistical AI search systems consistently evaluate as high-confidence over time. That is what makes ranking durable. That is what the volatility data, read correctly, is asking for.

Run the free analysis to establish your AI search ranking baseline — mention rate and position across platforms, results in 60 seconds.

Audit Your Website – Free

References

Aggarwal, P., Maatouk, A., Maillard, Q., Gagnon, L., Pal, C., & Boussioux, L. (2024). GEO: Generative engine optimization. Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ’24). https://doi.org/10.1145/3637528.3671900

Iyappan, S. K. (2026). From keywords to intelligence: A comparative framework analysis of SEO, AEO, and GEO in AI-driven digital ecosystems. GOYBO International Journal of Marketing Intelligence, 1(1), 1–20. https://doi.org/10.5281/zenodo.20362080

Kargaev, D. (2026). The SEO-to-GEO gap: Quantifying ranking factor divergence between traditional and generative search. SSRN. https://doi.org/10.2139/ssrn.6476021

Luther, V., & Touboul-Cohen, O. (2026). Brand visibility in AI search: A longitudinal analysis of AI visibility metrics in the U.S. tea industry. Whitebox / Boston University.

Published by AIO Clicks — Digital Visibility Specialists | Haaksbergen, Netherlands | aioclicks.com