Methodology
Where does every number come from, and what is it allowed to say?
Audience Lens combines four sources with different measurement logic. They are never merged into a single figure. This page is the contract: what each source is authoritative for, and what it must never be used to claim.
Import status
Which sources are carrying real exports today?
Comscore
IllustrativeNo dataset imported. Components fall back to illustrative values and are labelled as such.
Parse.ly
IllustrativeNo dataset imported. Components fall back to illustrative values and are labelled as such.
Google Analytics 4
IllustrativeNo dataset imported. Components fall back to illustrative values and are labelled as such.
Reader Voice interviews
IllustrativeNo dataset imported. Components fall back to illustrative values and are labelled as such.
Exports are mapped and validated in the import area.
Methodology. Every figure in the product is illustrative until the matching export is imported. This table is generated from the local dataset store, not authored.
Source authority
Which source wins when two of them disagree?
Comscore
ComscoreModelledMonthly, published with a two to three week lag
Authoritative for
- Audience size and reach in the US digital market
- Demographic, professional and geographic composition
- Indexes against the US internet population
- Competitive comparison, cross-visitation and duplication
Never use for
- On-site content performance
- Engaged time on Newsweek properties
- Anything requiring article-level detail
Panel-plus-tag hybrid measurement. Figures are modelled estimates and will not reconcile with site analytics.
Parse.ly
Parse.lyObservedContinuous, effectively real time
Authoritative for
- Pageviews, visitors and engaged time on Newsweek properties
- Article, author, section, topic and format performance
- Recirculation, return behaviour and content depth
- Entry referrer and first-read behaviour
Never use for
- Total addressable audience or market reach
- Demographics
- Campaign attribution beyond entry referrer
Tag-based observed measurement of Newsweek properties only.
Google Analytics 4
GA4ObservedDaily, subject to consent and modelling
Authoritative for
- Campaign and UTM attribution
- Landing page and conversion event tracking
- Retention cohorts and cross-session journeys
Never use for
- Redefining behavioural segments already defined in Parse.ly
- Any figure summed with Parse.ly visitors
Held for phase two. Event naming, campaign tagging and consent configuration must be validated before internal publication.
Reader Voice interviews
Reader VoiceQualitativeWeekly, ongoing
Authoritative for
- Why readers read, return, subscribe or leave
- Brand perception in readers' own language
- Unmet needs and product requests
Never use for
- Any statement of how many readers do or believe something
- Sizing an audience or segment
- Projection to the total audience
Small-sample directional research. Twelve interviews to date.
Comscore and Parse.ly will never reconcile. Comscore models people in a market; Parse.ly observes browsers on our properties. Any attempt to add or divide one by the other produces a number that means nothing.
Methodology. If a chart in this product shows a number, it carries a source chip. If two sources cover the same ground, the one listed as authoritative here is the one the product reports.
Data dictionary
What exactly does each metric mean, and where does it mislead?
| Metric | Source | Definition | Caveat |
|---|---|---|---|
| Unique audience | Comscore | Estimated unduplicated US individuals who visited a Newsweek property in the month. | Modelled. Not comparable with Parse.ly visitors and must never be summed with them. |
| Reach | Comscore | Unique audience as a percentage of the total US digital population. | Denominator changes with Comscore's population estimate; year-on-year moves can be partly definitional. |
| Index | Comscore | Composition of the Newsweek audience on an attribute divided by the composition of the US internet population, times 100. | An index of 168 means over-representation, not size. Always read alongside composition. |
| Visitors | Parse.ly | Distinct browsers observed by the Parse.ly tracker in the period. | Cookie churn inflates this figure; treat month-on-month moves under 3% as noise. |
| Average engaged time | Parse.ly | Mean seconds of active attention per pageview, measured by scroll, focus and interaction heartbeats. | Not comparable with GA4 average engagement time, which uses a different definition. |
| Recirculation | Parse.ly | Share of article views that lead to another Newsweek pageview in the same visit. | Sensitive to on-page module changes; annotate deploys before reading trends. |
| Return rate | Parse.ly | Share of a topic's visitors who visit Newsweek again within 30 days. | Attribution to a topic is by first read in the window and does not imply causation. |
| Loyalty half-life | Derived | Days until half of a topic's new readers have stopped returning. Derived from Parse.ly return curves. | Derived metric, not a native Parse.ly measure. Methodology needs review before external use. |
| Co-read index | Derived | Observed share of readers consuming two topics divided by the share expected if reading were independent, times 100. | Below 100 means the two topics are read together less than chance would predict. |
| Theme frequency | Reader Voice | Number of distinct interviews in which a theme was expressed at least once. | A count of interviews, never a percentage of the audience. Fewer than five interviews is flagged as low sample. |
Methodology. Every caveat here is shown because it has already caused a misreading in an internal report at some point.
Segment logic
How is a behavioural segment defined, and what still needs validating?
New visitors
46.2% of visitorsFirst observed visit within the selected period.
1 visits/month · 51s engaged · 9.8% recirculation
Validation needed
Cookie churn inflates this group; needs cross-check against a device-stable identifier.
Casual readers
22.4% of visitorsOne visit in 30 days, two or fewer articles.
1 visits/month · 58s engaged · 12.4% recirculation
Validation needed
Article-count threshold is a placeholder pending a real distribution review.
Returning readers
16.1% of visitorsTwo to three visits in 30 days.
2.4 visits/month · 79s engaged · 21.7% recirculation
Validation needed
Overlaps with engaged group at the boundary; confirm rounding rules.
Engaged readers
8.9% of visitorsFour to seven visits in 30 days, or average engaged time of three minutes or more.
5.2 visits/month · 138s engaged · 29.6% recirculation
Validation needed
The OR condition double-counts short-but-frequent and long-but-rare readers.
Highly engaged readers
4.3% of visitorsEight to fourteen visits in 30 days with recirculation above the site median.
10.4 visits/month · 176s engaged · 38.2% recirculation
Validation needed
Median recirculation moves month to month; consider a fixed threshold.
Loyal readers
2.1% of visitorsFifteen or more visits across three or more distinct weeks.
21.8 visits/month · 194s engaged · 44.1% recirculation
Validation needed
Small group; confirm bot and internal traffic are fully excluded.
Definition. Segments are behavioural, defined by observed visit frequency and depth. They are not demographic and not personas.
Methodology. Each segment carries the specific validation it still needs. Publishing that gap is deliberate: a segment used without knowing its weakness is more dangerous than no segment.
Research governance
How is reader interview data handled, and what protects participants?
Anonymize at intake
Names, employers and any identifying detail are stripped before an interview enters the repository. Only the anonymized descriptor, cohort and coarse attributes are stored here.
Recordings stay out of the product
Audio and video remain in the restricted research store with defined retention. Audience Lens links to the transcript record, never to raw media.
Evidence is typed, always
Every item is labelled as a direct quote, researcher observation, AI summary, recurring theme or hypothesis. Summaries are never presented as quotations.
AI output is draft until reviewed
Machine-extracted quotes and themes stay in a draft state and require researcher approval. Every edit is attributed and timestamped.
Small samples are flagged
Any theme supported by fewer than five interviews carries a low-sample marker wherever it appears.
No qualitative projection
Interview findings are never expressed as percentages of the audience and never used to size a segment.
Sensitive disclosures are excluded
Health, financial or political disclosures that could identify a reader are removed from the quote bank even when anonymized.
Internal use only
Quotations may be used in internal planning. External or marketing use requires a separate consent check against the interview record.
Methodology. These rules are enforced at intake, not at publication. Nothing identifying reaches this product in the first place.
Future data structure
How does a raw interview become structured, reusable audience intelligence?
Upload recording or transcript
Manual upload to the research store, transcript pasted into the interview record.
Drop a file into Audience Lens; transcription runs automatically with speaker separation.
Create anonymized profile
Researcher fills in cohort, reader type, age range, region, industry and platforms by hand.
Draft profile pre-filled from the transcript with identifying detail auto-redacted for review.
Extract memorable quotes
Researcher selects quotes and records the transcript timestamp.
Candidate quotes proposed with timestamps; researcher approves, edits or rejects each one.
Identify recurring themes
Themes assigned from a controlled vocabulary maintained by the research lead.
Clustering proposes new themes and flags when an existing theme should be split or merged.
Tag needs, motivations and barriers
Manual tagging against the reader-need taxonomy.
Suggested tags with confidence, always requiring human confirmation.
Generate jobs-to-be-done
Written by the researcher from the interview summary.
Drafted from clustered evidence and reviewed before publication.
Connect to segments and personas
Researcher links the interview to the personas it informs.
Persona evidence stacks update automatically, with confidence recalculated from interview count and consistency.
Track theme movement
Monthly mention counts maintained in the theme record.
Automatic trend tracking with alerts when a theme accelerates or fades.
Every record in this product already carries source, data type, period and last updated. That is what allows mock data to be swapped for real exports without redesigning a single view.
Methodology. Column one is today's manual process. Column two is the assisted pipeline this data model is designed to support. Human approval stays in every step.
Known gaps in this release
What should nobody try to answer with this product yet?
No subscriber state
No source in this release identifies whether a reader is a subscriber. Every subscription finding here is qualitative and cannot be sized.
No cross-session journeys
Parse.ly shows entry referrer and in-visit paths. Multi-visit journeys require validated GA4 and are out of scope.
No revenue or yield data
Advertising and subscription revenue are not connected. Audience value statements here are behavioural proxies, not commercial ones.
Qualitative sample is small and skewed
Twelve interviews weighted toward paid and formerly paid readers. Acquisition findings are weaker than retention findings.
Loyalty half-life is a derived metric
Useful for ranking topics against each other. Not a validated absolute measure and not for external use.
GA4 variance is unexplained
GA4 consistently reports below Parse.ly. Until the cause is separated into consent, definition and tagging effects, GA4 stays out of the product.
GA4 validation queue before phase two
Registration event naming
Not validatedConfirm a single canonical event fires on account creation across web and app, with no duplicate legacy names.
Subscription conversion events
Not validatedConfirm purchase or subscribe events fire server-side and are not blocked by consent tooling.
Campaign tagging discipline
Partially validatedAudit UTM conventions across newsletter, social and paid so channel grouping is trustworthy.
Consent mode configuration
Not validatedEstablish how much traffic is modelled rather than observed before any GA4 number is published internally.
User properties
Not validatedDetermine whether any property legitimately exposes subscriber status. Until then, no subscriber segment exists.
Methodology. Stated openly so findings are not stretched beyond what the data supports.