ChatGPT
Employer naming and descriptions are retained separately from evidence-retrieval checks.
BrandScore methodology
How BrandScore Measures AI Employer Brand Intelligence
BrandScore measures whether employers are named in open candidate answers, what AI says when directly asked about an employer, and which sources and passages can be inspected. These behaviours remain separate.
BrandScore does not replace your employer brand strategy. It records observed descriptions and evidence for your team and agency to inspect; those descriptions do not certify employee experience.
NEW
NEWCONFIGURATION AND SCALE
Planned executions, not guaranteed answers.
The standard 117-question set across six experiences, sampled twice, plans 1,404 executions. This includes both open-market and directed employer questions. Report types, adaptive sampling and market scope can change the total; additional retrieval and analysis work is separate.
Scheduled monitoring runs every 14 days. Seven total assessments form the 90-day programme; Annual Monitor provides 26 scheduled assessments plus five bespoke role/skills studies monthly. A bespoke study has a separate scope and is not promised 1,404 executions.
01 · Candidate-style questions
Structured around how candidates research employers
BrandScore uses a structured question set modelled on the ways candidates explore potential employers. The same question framework supports consistent comparison between employers and across measurement periods.
The methodology publishes the subjects measured, not BrandScore's proprietary prompts or scoring logic.
01
Which employers should candidates consider?
02
Does an employer match a specific Candidate Priority?
03
What may working for an employer be like?
04
How do employers compare?
05
Is an employer named in an open candidate answer?
06
What evidence supports the answer?
02 · Six AI experiences
Each engine provides a different view
BrandScore measures ChatGPT, Gemini, Claude, Perplexity, Google AI Overviews and Google AI Mode. Each can produce different employer mentions, narratives, sentiment and sources, so measuring one engine does not provide a complete view.
Gemini
An independent view of employer naming and direct-employer descriptions.
Claude
Measured for Answer Presence and narrative; contrast-only, not scored for AI Readiness.
Google AI Overviews
Real Google Search answer captures; a missing Overview is distinct from a technical failure. Not used for AI Readiness.
Google AI Mode
Real AI-powered Google Search answers, not Gemini API simulations. Not used for AI Readiness.
Perplexity
Search-enabled in the current visibility method; citations still require claim-level inspection.
03 · Separate evidence types
Open-market and employer-named measurement
BrandScore keeps two types of candidate question separate so answer presence is not confused with employer narrative or sentiment.
Open-market questions
Measure whether an employer is named when candidates ask open questions without the employer's name. This evidence supports AI Answer Presence, competitor mentions and comparative visibility. Presence is not endorsement.
Employer-named questions
Measure what an AI engine says when asked directly about a named employer. This evidence supports narrative, tone, conviction and Candidate Priority analysis when enough eligible answers are available.
04 · Competitive benchmarking
Comparison with employers competing for the same talent
BrandScore compares an employer with organisations that appear in the same candidate-style answers or compete for the same talent. It is an employer and talent-market comparison, not a generic company or consumer-brand benchmark.
Market Maps select the most visible employers in the measured market before presenting eligible favourability. Additional named employers remain in the wider analysis without unsupported favourability positions. Five selected employers receive deeper comparison; they do not limit open-market discovery. Answer position describes where a name appears inside an answer, not an independent employer ranking. Head-to-head comparisons need sufficient qualifying shared answers.
Answer Presence
Open-question employer naming
AI Favourability
Candidate Priority associations
Talent competitor mentions
Market Maps
Narrative differences
05 · Narrative and sources
What AI says, and the evidence visible around it
BrandScore examines tone, conviction, recurring associations, Candidate Priority alignment, sources and citations, and differences between an employer's intended EVP and the narrative produced by AI.
A citation shows a source associated with an answer. It does not automatically support a claim. Inspect the relevant passage, exact employer attribution and limitations. A failed retrieval or technical check means we could not verify the material, never that evidence is absent or AI cannot read it.
Narrative
Tone, conviction and recurring associations
Alignment
Candidate Priorities and intended EVP
Evidence
Visible sources and citations
CITATIONS ARE NOT VERIFICATION
Three distinct evidence stages.
1. Citation returned
The AI experience returned a source URL or citation. Source mentions in prose are not automatically genuine citations.
2. Page readable
Eligible evidence workflows check retained page content or retrieval. A stored link does not prove that a page was readable, or read live by an AI experience.
3. Claim support checked
Eligible Readiness and evidence-validation workflows assess relevant passages and employer attribution. Ordinary citation analysis does not verify every claim on every cited page.
These checks have different scopes. A citation is not proof that a source caused an answer. An unsuccessful retrieval or unsupported passage remains an explicit limitation, not proof of absence.
AI READINESS ELIGIBILITY
Not six identical retrieval tests.
GPT-5, Gemini and Perplexity are the scored services. Claude provides contrast-only observations; Google AI Overviews and AI Mode do not contribute to Readiness scoring. Technical accessibility and structure checks are separate from evidence retrieval.
Usable memory-only answers and live-retrieval answers are distinguished. Memory answers do not prove a page was read live. Usable coverage must reach at least 90% combined across scored services; per-service coverage remains visible. This is an evidence-coverage requirement, not a promise that 90% of answers are favourable or verified live.
Failed, cut-off, misattributed or ungraded evidence is not interpreted as neutral tone or employer absence. Required unfinished work blocks a completed result; retained excerpts must be attributable to saved answers.
06 · Repeated measurement
Patterns matter more than one-off self-testing
AI answers can vary between runs. Engines use different sources and retrieval methods, and results can change as source material and models change. A single self-test cannot distinguish ordinary answer variation from a sustained pattern.
BrandScore uses consistent questions and repeated samples, then measures on the product's verified fortnightly cadence. Compare only compatible question populations, markets and methods; model or configuration changes can limit historical comparability. This creates comparable assessment periods and reveals patterns over time without suggesting that any result is permanent.
Fortnightly cadence
1
Measure
2
Compare
3
Observe change
35,529
genuine citations
3,001
domains
133
employer assessments
As of August 2026, BrandScore had analysed 35,529 genuine citations across 3,001 domains from 133 employer assessments. These figures describe historical four-engine research, not current six-experience results, customer numbers or total market coverage.
07 · Quality and interpretation
Observed behaviour, interpreted with care
BrandScore reports observed AI behaviour. It does not claim to control an engine, predict every future answer or turn a visibility signal into an endorsement claim.
- Results represent measured answers during the assessment period.
- AI outputs can change as models, retrieval methods and source material change.
- Visibility does not equal endorsement.
- Being mentioned does not automatically mean being recommended.
- Favourability and narrative are reported only when enough eligible evidence is available.
See how AI represents your employer brand
Measure whether AI names your employer, inspect descriptions and sources, and compare observed findings with employers competing for the same talent. AI Answer Presence is naming, not endorsement.
