Follow
Follow The Observatory
A recurring public record of how brands and market actors appear across AI answers and search — what changes, what remains stable, and what the evidence helps an organization understand before it acts.
Quick answer
The Observatory is an independent publication tracking how brands and other market actors appear in AI answers and search across covered premium markets. It publishes recurring, governed readings of organic rankings, answer openings, citations, framing and differences between named models. Deliberately designed query panels and preserved observations build a record beyond one prompt. Different readings answer different questions: who appears, which sources support an answer, and how a market is described. The purpose is to understand that public representation before drawing conclusions or choosing an action. Each reading states the evidence and limits behind it. — The Observatory
At a glance — key facts
- One market can have several different first places. An organic leader, an AI opening actor and a leading cited source perform different roles. Combining them into one “AI ranking” hides useful evidence.
- The question is part of the result. A market panel typically contains around one hundred human-validated queries; Query Spotlight examines one exact flagship question within that wider view.
- History and variation belong together. Repeated observations reveal what recurs and what differs by question, surface, model or period. A claim of change needs comparable evidence.
- Observation comes before intervention. A visible gap can identify the question an organization needs to investigate. It does not establish the cause or prescribe the remedy.
- Start with the evidence that matches your question. Choose a market and destination, then check the reading’s surface, window and Reading Basis before comparing positions.
Search has moved beyond the list of links
Traditional search visibility is easy to picture: a page appears at a position for a query, a user clicks it, and analytics records what happens next. That system still matters. AI answers add another public layer in which brands are named, sources are cited and markets are described before a reader reaches a website.
A company may be prominent in search results yet receive little attention in a generated answer. Another may be mentioned repeatedly but described as an intermediary rather than as the provider a customer is seeking. A publisher may supply the cited evidence for an answer about businesses it does not operate.
Presence, citation and framing are three different things. Presence establishes that an actor appears in the observed output. Citation establishes an explicit source relationship. Framing describes the role assigned to the actor: a provider, platform, publisher, example or reference, among other possibilities. Opening prominence adds another question: which actor or source does the answer open with?
Visibility is not authority — being seen is not being chosen. Here, authority means being treated as a reference on a defined surface. “Chosen” does not mean that a customer selected the business or that the AI recommended buying from it. The visibility–authority distinction keeps presence separate from the citation and framing evidence needed to describe reference treatment.
The Observatory records these relationships separately. The distinction becomes clearer when the readings name different actors in the same market.
What does “first” mean in an AI-search reading?
In one matched Paris luxury-hospitality comparison, Yonder, a publisher, is first in organic search; Atout France, an institution, is first in AI Overview openings; and YouTube, a citation host, is first in AI citations. These are three answers to three different measurement questions.
Paris luxury hospitality, Google desktop in France, September 14–20, 2026. “First” below means first in the published period table, not first in every individual response.
| Public reading | First-ranked label | What orders the reading |
|---|---|---|
| Organic Ranking — Top 5 | Yonder (yonder.fr) | Frequency of top organic position |
| AI Overview Ranking — Top 10 | Atout France (atout-france.fr) | AI Overview opening share, derived from observed opening volume |
| AI Citation Ranking — Top 5 | YouTube (youtube.com) | Cumulative AI Overview citation volume |
A publisher, an institution and a host can each lead a different reading of the same market. The host’s domain does not identify every underlying video author or the hotel being discussed. A brand’s search position cannot answer whether it opens an AI answer, and the opening cannot answer whose material is cited.
This is a bounded example, not a permanent Paris hierarchy. The Paris evidence synthesis follows each reading separately over time. One surface cannot substitute for another; this is why The Observatory separates them.
“A search position, an answer opening and a cited source describe different roles in a market’s public representation.”
The Observatory
How does The Observatory observe a market?
A market and its questions are deliberately defined
An observed market combines one vertical with one destination: private banking in London, for example, or luxury fashion in New York. The query panel is designed to represent different ways people approach that market. It is not a bag of random prompts or a list tailored to make one company appear.
The panel is built in three human-validated layers. A structured baseline begins with roughly 80 queries covering the market’s activities and intents. Long-tail research then develops a wider candidate list, from which approximately 20–25 questions are curated. Generic completion restores the shorter, everyday wording that research alone can underrepresent. Review and curation determine the enabled panel; these layers are not a promise to add every candidate together.
A full market panel typically contains around one hundred governed queries. Panel depth varies by market; Method documents the current median panel size and the Reading Basis states the coverage applicable to each reading. The baseline of roughly 80 is therefore not the total panel.
The questions cover discovery, commercial needs, experience, comparison, trust, authority and particular market segments. Local place names and micro-locations sit alongside wording that does not repeat the destination, because the capture itself has a geographic scope. Language is also recorded: an English run and a French run are separate observations. Designated flagship questions provide the focused view used by Query Spotlight.
Each layer passes human review. The public description explains the panel’s construction without publishing the full query universe or private prompt libraries. For the detailed panel, language and activation rules, Method’s market-panel explanation is the deeper layer.
Repeated runs create the period reading
For period-based market readings, the governed observation cycle runs from Monday through Sunday, with weekly aggregation at the close of that cycle. Period readings bring together comparable observations within the stated window. The observation window is distinct from the later publication date.
A panel of around one hundred queries, observed repeatedly, can yield hundreds of query-level observations in a weekly reading. The relevant total is the evidence actually available: the reading’s query scope, comparable run count and window. Missing runs, unavailable AI outputs or a different panel can change that coverage. A one-query Spotlight and a single narrative capture do not inherit the volume of a full market panel.
This is why the Reading Basis matters. It tells the reader what supports the published result, rather than asking them to infer coverage from a weekly label. Method & Standards explains how that scope and evidence are reported.
Which public reading answers which question?
Organic Ranking — Top 5: who appears strongly in conventional search?
Organic Ranking orders the top five actors by top organic position across the observed period. Organic coverage describes the share of observed queries on which an actor appeared; Presence describes the comparable runs in which it was observed. The reading concerns the organic search surface. A first organic position is neither an automatic first place in an AI answer nor a measure of commercial leadership.
AI Overview Ranking — Top 10: who opens the observed AI answer?
AI Overview Ranking orders ten actors by their share of observed Google AI Overview openings over the period. The opening is identified under the applicable capture rule; it is not simply the first link in a citation panel. An actor can be mentioned elsewhere in an answer without opening it. Opening share therefore answers a more specific question than “was this brand visible?”
AI Citation Ranking — Top 5: which sources are visibly cited?
AI Citation Ranking orders five actors by cumulative citation volume in the observed Google AI Overviews. Citation share describes their share of the recorded citations; Presence identifies comparable runs containing citation evidence.
A citation can point to a brand’s own page or to a third-party page discussing it. Those are different source relationships. Citation is not recommendation, and a hosting domain is not necessarily the actor being discussed. Visible citations also do not reveal every source used internally to produce the answer. The Citation Surface reference explains how mention, citation, opening and explicit recommendation differ.
Query Spotlight: what happens when the customer asks one precise question?
Query Spotlight isolates one exact flagship query from the wider market panel and observes that question across comparable runs. Its measured surface is Google organic search. It is not a generic label for a ChatGPT answer or an AI Overview.
The New York luxury-fashion record shows why this narrower view matters. In a matched window, the aggregate organic top five lists Reddit, Yelp, Instagram, NYC Tourism and Vogue. For the exact question “luxury shopping districts New York”, the Query Spotlight records frommers.com at median best organic position 1.
Matched evidence: aggregate organic selection and exact-query Spotlight, June 15–21, 2026.
The domain is visible in the precise-query reading without appearing in that period’s aggregate top-five selection. It was not therefore absent from all underlying market observations. Nor does median position 1 mean first place in every run: Query position summarizes the runs in which the actor appeared, while Presence reports how many comparable runs included it.
The lesson is that a broad market selection cannot answer every customer question. Changing from an aggregate view to an exact query reveals a different relationship; the comparison does not isolate query wording as a causal explanation.
DNC Lens: how does a named model describe the market?
DNC Lens publishes a scoped reading of one named model’s recorded account, produced through governed Deep Narrative Capture. Where a ranking describes position or presence, a Lens examines how the model represents the market: which actors it foregrounds, what roles it assigns and how it describes their significance.
The captured response is preserved verbatim. Public interpretation can select and contextualize that response while keeping attribution clear. A model narrative is evidence of what the model said; it is not verification that the description is true. The Deep Narrative Capture reference explains the narrative layer and its publication boundaries.
Model Watch: where do model accounts converge or differ?
Model Watch compares the named model readings of the same market. It makes agreement and disagreement visible, including different actor emphasis, source roles and framing. Sometimes the difference is more fundamental: the accounts appear to organize the market around different kinds of participants.
In a recorded New York luxury-fashion pair, GPT-4o native names Chanel, Gucci and Louis Vuitton, while Claude Sonnet 4.6 native names Vogue, WWD and CFDA. These selected names bring fashion houses into one account and publications and an industry organization into the other.
Capture attribution: September 21, 2026 — Model Watch comparison, GPT-4o native Lens and Claude Sonnet 4.6 native Lens. These are examples from the accounts, not exclusive inventories.
The useful question is no longer only “which brand ranks higher?” It is also “what does this model seem to be describing when it describes the market?” The pair shows a difference in recorded representation. It does not establish that one provider is better, expose hidden reasoning or verify the models’ descriptions of those organizations.
Model, version, surface and mode remain part of the attribution. A native-versus-web difference is not automatically a provider difference. Measured rankings and model narratives remain separate evidence layers, even when reading them together is informative.
What can a gap between organic and AI presence tell us?
Strong organic presence can coexist with limited AI-answer presence. Frequent citation can coexist with little opening prominence. Stable presence can coexist with different framing. These divergences identify different questions for investigation.
A cross-surface comparison of observed organic presence with AI-answer presence over the same window identifies differences in where an actor appears. It does not by itself measure quality, authority or risk.
Such a comparison differs from the conceptual Authority Gap: strong visibility without corresponding reference treatment on the same scoped surface and window. Cross-surface presence and same-surface reference treatment should not be substituted for each other. The AI Visibility vs AI Authority Guide explains how to interpret the distinction. Neither kind of gap establishes its own cause.
A single prompt is a snapshot. A recurring reading creates a record.
Anyone can ask an AI system a question and record the answer. That can be useful. The harder question is whether the observation persists, disappears or changes under comparable conditions.
Preserved history supplies the time dimension. Multiple queries, runs, surfaces and model readings reveal variation within the record. Together, they help distinguish a recurring appearance from a one-off result, a precise-query exception from a broad selection, and a framing difference from a measured change in position.
Repeated dates alone do not establish comparability. The market, query panel, measure, language, location, device and relevant model or surface conditions must support the comparison. A changed panel can alter who appears; a missing output can reduce the available evidence. Neither should be silently turned into an actor’s rise or disappearance.
Where sufficient comparable history exists, a reading can show measured representation holding, rising, weakening or no longer being recorded within scope. Recurring membership alone is not a growth trend. Month-to-month or quarter-to-quarter conclusions need evidence that supports that longer comparison. Method’s comparability guidance sets out those requirements.
SEO still matters. Understanding the representation comes first.
SEO remains important to how an organization is found and understood in search. Google’s guidance for AI Overviews and AI Mode explicitly retains foundational SEO practices and states that no additional requirements apply to appearing in those features. That guidance does not guarantee a particular brand’s inclusion or explain an observed selection.
For a brand, organic search is one layer of a larger representation question. Does the organization appear in AI answers? Does it open the answer? Is its own material cited, or is another source discussing it? Is it represented in the intended role? Does a precise customer question bring it forward? Does another model describe the market differently?
The Paris example separates a search-position question from an opening or sourcing question. The New York Spotlight separates broad selection from a specific shopping task. The model pair adds a question about how the market itself is framed. These distinctions help identify the type of issue before anyone chooses what to publish, optimize, structure or change.
The order is to understand and observe, then diagnose, then act. Public evidence can establish the observed pattern and help define the investigation. A brand-specific explanation of its cause, a remediation plan or a claim that an intervention worked requires additional evidence and deeper diagnosis.
How can SEO, GEO and AEO agencies use the record?
Agencies and brands can use a neutral public evidence layer to make the client’s question more precise. “Do we rank?” may concern organic position, AI presence, citations, opening prominence or framing. “Are competitors recommended?” requires explicit recommendation evidence; more mentions or citations alone do not answer it.
Scoped Observatory readings can give agencies, analysts and their clients a shared basis for discussing what was observed. They can also reveal that a prominent “competitor” is a publisher, platform or other source rather than a commercial rival. The competitor-visibility Guide explains the comparison needed before drawing a conclusion. Deep actor-specific diagnosis remains private.
Can the record show what changed while SEO or GEO work was underway?
Yes, where the evidence is sufficient and comparable. Readings can show whether measured organic position, AI presence, citation, opening prominence or framing changed during the same period as an organization’s work.
A measured change does not identify its cause. No detected change does not prove the work had no effect outside the observed scope or period. The first question is factual: did the measured public representation change? Explaining why requires additional evidence.
Governed and auditable by design
A reading becomes useful evidence when a reader can identify what was actually observed. Its scope, observation window and Reading Basis establish the relevant surface, queries and comparable runs. Source and capture provenance connect the published claim to its evidence where applicable.
The Attestation Standard organizes that accountability through Scope, Basis, Window, Source and Limitation. Auditability supports inspection of the published evidence trail. It does not imply independent assurance of every output, public access to private calculations or a guarantee that another reader can reproduce a changing system’s answer.
Historical readings are preserved rather than silently overwritten because a newer observation exists. This makes it possible to distinguish a current observation, recurrence, supported change and a conclusion the history cannot yet sustain. Method’s standards explain the reporting requirements; the Citation library provides reuse material for eligible published observations with its attribution, scope and limitations.
Where AAI Kernel fits
AAI Kernel is The Observatory’s founding intelligence source and supplies much of the governed intelligence published here. AAI Kernel produces the governed intelligence; The Observatory turns selected findings into independent public readings.
The public record is a selected market-level view of a deeper governed intelligence capability. The wider infrastructure can support private market and actor diagnostics and partner use. Private client evidence never becomes source evidence in public readings. The market-centric and client-centric distinction explains that firewall.
The Observatory may also incorporate other qualified authoritative evidence when its publication and evidence rules permit it. The source and editorial independence policy explains the relationship. Public positions are not sold.
Questions about brands in AI answers and search
How does AI rank brands?
There is no single universal AI ranking. Organic position, AI-answer presence, opening prominence, citation and framing describe different observations on named surfaces. The Observatory records them separately within a defined market and period. A ranking describes the observed result; the ranking interpretation Guide explains what its scope and supporting signals allow a reader to conclude.
How do I know what ChatGPT or another AI says about my company?
Start with the relevant market and destination, then find an available reading for the question and named product, model or surface. Inspect the recorded account, excerpts or citation evidence and its date. A Google result cannot establish what ChatGPT said. Published market coverage may not include your company or every question; unavailable coverage cannot establish absence.
Why is my brand missing from AI answers?
A brand missing from a published top-five or top-ten selection may still occur in the underlying outputs. First distinguish selection limits, missing capture and usable answers in which the identified brand was not observed. A supported absence remains specific to its query, surface and window. The brand-absence Guide explains those checks before any cause is proposed.
Why do competitors appear before us or more often?
The record can establish a difference in position or frequency when the entities, query mix, measure and observation conditions are comparable. It cannot derive the cause from the difference alone. The other actor may also be a source or intermediary rather than a direct business rival. The competitor-visibility Guide explains how to make that distinction.
How do I know whether AI recommends my company?
Look for explicit language or context presenting the company as a choice to consider in the recorded answer. A mention, opening position or citation alone is insufficient. Even an explicit recommendation is evidence of that answer’s wording, not The Observatory’s endorsement or proof of service quality. The Citation Surface reference separates these roles.
How is AI visibility different from SEO visibility?
Organic search visibility concerns appearance on the conventional search surface. AI visibility concerns appearance in the named generated-answer surface. Citation and framing add evidence about how the answer uses or describes an actor. A strong organic position does not guarantee AI presence or reference treatment. The AI Visibility vs AI Authority Guide explains how to interpret those signals.
Can SEO, GEO and AEO agencies use Observatory intelligence?
Yes. Agencies can use available public readings as scoped evidence when discussing search and AI representation with clients. The reading’s surface, query scope, window, attribution and limits must travel with the claim. Public observations can help define a diagnostic question; they do not supply a client-specific cause, action plan or proof that an agency’s work produced a change.
Can The Observatory show whether visibility is changing?
Where sufficient comparable history exists, it can show changes in the measured position, presence, citation or other reported signal. The measure determines what “improvement” means. Repeated inclusion alone does not establish a trend, and a later date does not make two readings comparable. Method’s comparison guidance explains what a historical claim requires.
Why not just ask ChatGPT manually every week?
Manual prompting can produce a useful record when collection conditions, coverage and evidence are documented consistently. Repeating a question alone does not establish a comparable trend. The Observatory adds recurring publication of scoped readings, multiple observation dimensions, preserved history and evidence lineage. A model’s own explanation of why it named a brand is not an internal reasoning trace.
Where does The Observatory’s intelligence come from?
Much of the governed intelligence comes from AAI Kernel’s measurement infrastructure. The Observatory publishes independently and may use other qualified authoritative evidence under its publication standards. Public readings remain market-centric; private client evidence does not enter their source record. The source policy explains that separation.
Evidence basis and limits
The examples illustrate how to read the evidence, rather than establish today’s order. The Paris comparison uses three matched period readings from September 14–20, 2026. The New York query example uses the June 15–21, 2026 aggregate and exact-query Google organic readings. The New York narrative example uses the September 21, 2026 GPT-4o native and Claude Sonnet 4.6 native pair. Each linked publication remains the source for its observation window, scope and evidence.
The public record describes external outputs, not model weights or internal ranking systems. The three examples have different measures and coverage. They are not a combined score, a cross-market league table or a count of independent experiments. Recorded model descriptions remain attributed perceptions. Observed differences do not establish hidden system reasoning, an actor-specific cause, commercial impact, product or service quality, endorsement or recommendation.
The Observatory does not claim to measure real-world revenue, market share, or business leadership unless explicitly stated. It publishes observed AI and search positioning within a defined scope and time window.
Intelligence by AAI Kernel. Published by The Observatory.
Continue the reading
Once the observation question is clear, the market and destination establish which public record applies, and the dated rankings show the available measured positions. Those positions become interpretable through their scope and denominator, which Method & Standards explains. If the question is whether appearing also means being treated as a reference, the AI Visibility vs AI Authority Guide explains the additional citation and framing evidence needed before reaching that conclusion.
