01 / SCAN
When are clients recommended?
Sample real buyer questions to capture recommendations, omissions, competing brands, and recurring citations.
Evidence-based AI recommendation intelligence
Know when AI recommends your clients and what to do about it.
BrandScanner examines when AI recommends or overlooks a client, verifies the evidence behind those answers, and identifies competitive gaps worth investigating. Turn the findings into defensible recommendations and practical next steps.
Built for SEO agencies, GEO consultants, and independent search strategists.
Repeatable AI recommendation samplingSource-checked evidence and uncertaintyPrioritized recommendations for client action
Initial coverage: ChatGPT + Gemini · Agency and independent consultant pilots first · Transparent methodology
SCAN Illustrative · sample summary
BUYER QUESTION“best accounting firm in boston”
Repeated answer sampling100 responses per assistant
ChatGPTRecommendation frequency
Competitor A recommended71 / 100
Your brand recommended6 / 100
GeminiRecommendation frequency
Your brand recommended18 / 100
Your brand mentioned41 / 100Supporting measure
Mentions include neutral references and do not necessarily mean recommendations.
What deserves investigation?
Observed →In these sample results, competitor recommendations are more frequent on ChatGPT.
Next step →Check recurring sources and verify the evidence behind service claims.
Illustrative counts, not live client results. Explore the Intelligence Brief for Wilson intervals, source verification and prioritized actions. The 100-response sample is an example, not a validated minimum.
Understand the recommendationWho AI favors and where a client is absent.
Examine the evidenceWhich cited claims are actually supported.
Prioritize useful actionWhere investigation could improve client strategy.
The BrandScanner model
From AI answers to recommendations your clients can act on.
Visibility is a starting signal. BrandScanner is being designed to help agencies interpret what AI recommends, check the supporting evidence, and prioritize the gaps that deserve attention. Every finding distinguishes observation from inference.
02 / OBSERVE
How does AI compare them?
Find where competitors appear, which claims are emphasized, and what patterns repeat across sampled answers.
03 / VERIFY
What supports the answer?
Check cited sources against specific claims rather than assuming a citation automatically proves them.
04 / DIAGNOSE
What should happen next?
Identify possible source, entity, content, and competitive gaps; label hypotheses clearly and prioritize sensible follow-up.
For agency reporting: Make the observation clear, show the source evidence, explain uncertainty, and recommend the next investigation or improvement. Explore the example brief
EXAMPLE INTELLIGENCE BRIEF
Evidence your clients can see. Next steps your team can use.
Explore how sampled recommendation patterns can become a documented investigation, without claiming to know an AI model's private reasoning.
Client recommendation analysisExample category: professional services · Boston, MA
Illustrative dataBuyer question
“Best accounting firm in Boston”
Recommendation frequency · 100 sampled answers
Competitor A71 / 100
Illustrative 95% Wilson interval: 61.5%–79.0%
Your brand6 / 100
Illustrative 95% Wilson interval: 2.8%–12.5%
Your brand mentioned (secondary)41 / 100
Illustrative 95% Wilson interval: 31.9%–50.8%
These illustrative Wilson intervals assume independent responses. They do not represent population exposure.
What the evidence says
Observed Measured pattern
Supported Source checked
Inferred Hypothesis
Unknown Not verified
Observed
In this example, Competitor A appears more frequently in sampled ChatGPT recommendations.
Supported
Two recurring cited sources support their referenced claims; one offers partial support.
Inferred
Thinner independent evidence for your equivalent claim may contribute to the gap. That explanation remains a hypothesis.
i
Business-outcome check: No reliable association with branded-search movement detected in this illustrative 12-week window. This does not prove zero impact.
All brands, question patterns, counts and results in this demo are fictional examples. BrandScanner is in development; this is not a live client report.
Surface rollout
Focused coverage first. Broader coverage next.
BrandScanner will report results for each AI surface separately. Additional coverage will follow measured usefulness, user demand, and the ability to sample each experience appropriately.
Core recommendation scan
Build a baseline for how the initial assistants describe, compare and recommend client brands.
ChatGPTGemini
Search-native comparison
Examine recommendation and sourcing differences in a citation-forward answer engine.
Perplexity
Google AI search experiences
Measure AI Overviews and AI Mode separately from Gemini API outputs, using fit-for-purpose methods.
AI OverviewsAI Mode
Additional assistants
Expand when customer needs and evidence quality justify broader coverage.
CopilotGrokClaude
The roadmap describes intended coverage, not currently available integrations or equivalent data across different AI products.
Confidence and clarity
Show clients what the evidence says and what remains uncertain.
BrandScanner's proposed reporting standard separates measured results, verified support, reasonable hypotheses, and information that AI platforms do not reveal.
Observed
“A competitor appears more frequently in sampled answers.”
A pattern measured within a defined sample, with its surface and date range disclosed.
Supported
“This cited source supports the specific claim.”
The source was examined against the statement made in the AI answer.
Inferred
“Thin independent evidence may help explain the gap.”
A hypothesis for investigation, not a proven reason the model chose its answer.
Unknown
“The model's internal weighting of this source.”
Information the AI provider has not exposed and BrandScanner cannot verify.
Beyond a visibility score
Move from monitoring mentions to making better client decisions.
Counts can reveal a pattern. Context, verified support, and carefully labeled diagnostics help an agency decide what deserves attention.
Common visibility metrics
Useful descriptions of sampled answers.
- Mentions and placement
- Share of voice
- Sentiment
- Citation totals
- Competitor presence
BrandScanner recommendation intelligence
Evidence and context to support next steps.
- Recommendation and omission context
- Sources verified against specific claims
- Competitive displacement patterns
- Observed versus inferred diagnostic findings
- Prioritized opportunities to investigate
- Exploratory business-outcome signals where data permits
Built for client work
For agencies and independent SEO/GEO consultants alike.
Whether you manage a portfolio of client brands or advise a handful of businesses, BrandScanner is being designed to help you produce clear, checkable findings and useful next-step recommendations.
Client-by-client context
Organize buyer questions, brands, competitors, sampled answers, and supporting evidence around each client's market.
Transparent methodology
Disclose which assistants, prompts, time periods and samples support each result; identify unknowns explicitly.
Client-ready findings
Translate observed patterns and source checks into plain-language opportunities clients can understand and evaluate. White-label reporting is a planned capability.
For those who want to examine the evidence
The method and the hard questions, one level deeper.
The practical value comes first. Expand the sections below to inspect planned statistical methods, limitations, and the current research debate.
Measurement and methodology
Sampling, recommendation rates and confidence intervals
The proposed standard is repeated answer sampling for buyer-intent prompts, with model, prompt, region, dates and sample size shown in reports. A target floor of 100 answers per prompt is under evaluation for pilot economics and statistical usefulness.
Confidence intervals describe variation in the specified sample under the statistical assumptions used. They do not imply that prompts are representative of all customer questions or estimate the proportion of all AI users exposed to a recommendation.
Source verification and finding classifications
Brand mentions, recommendations, omissions and citations are classified separately. Cited pages are examined against individual claims and can be labeled supported, partially supported, insufficient, contradicted or irrelevant.
Diagnostic conclusions are independently marked observed, supported, inferred or unknown. BrandScanner does not claim access to proprietary ranking weights or hidden retrieval decisions.
Exploring business outcomes without overstating causality
With participating clients' permission, pilot analyses may compare recommendation patterns with branded search, engaged direct visits and self-reported AI discovery, using defined time windows and sensitivity checks.
These tests can surface associations or no detectable signal. They do not establish that AI recommendations caused traffic or revenue changes. Initial pilot analysis may be performed manually using GSC and GA4 exports.
Why Google AI search is a separate roadmap item
Gemini API answers and Google's AI Overviews or AI Mode are different experiences, with different exposures and measurement constraints. BrandScanner will not treat one as a proxy for the other or combine them into a misleading universal rank.
Research and buyer objections
Doesn't AI send very little website traffic?
Recent cross-site research finds much smaller direct AI-assistant referral volume than Google organic search. BrandScanner does not promise significant referral traffic or revenue from AI visibility alone.
Its aim is to document recommendation behavior, assess supporting evidence, help prioritize possible improvements, and explore whether relevant business signals change.
Can this prove why a competitor was recommended?
No. A scan may show recurring sources, competing claims and recommendation patterns. Internal ranking mechanisms are usually not observable. Explanations are labeled as hypotheses rather than causes.
Is 100 samples enough to make a marketing decision?
Not automatically. The useful sample size depends on variation, the size of the change being tested, prompt selection, dependence across responses and costs. Pilot work will validate sampling standards; the page does not promise certainty from any fixed number.
What if there is no measurable business effect?
A null result is useful. It helps an agency distinguish observable recommendation changes from claims of commercial return that have not been demonstrated. BrandScanner's reports should make this distinction explicit.
Founding partner pilot
Build a more defensible way to advise clients on AI recommendations.
We're inviting a small group of SEO agencies, independent GEO consultants, and search strategists to evaluate early measurement and diagnostic workflows with realistic client questions.
Explore how ChatGPT and Gemini represent client brandsReview source evidence and prioritized diagnostic opportunitiesHelp shape practical agency reporting and workflowsDiscuss scope, capacity and terms before joining