Methodology

What TalentTell measures

TalentTell analyses public employer brand communications and scores them across five dimensions. We assess your employer brand across four categories of content - your employer narrative, your stated identity and principles, how you showcase your people, and the consistency of your job adverts. The number of pages on your career site does not affect your score. What matters is the quality, specificity, and distinctiveness of what you communicate.

These dimensions measure how effectively an organisation communicates its employer brand, not how good the employer brand itself is. The distinction matters. A company could have a brilliant culture but communicate it poorly. TalentTell measures the communication, not the culture behind it.

The five scoring dimensions

Each dimension is scored from 0 to 100. Scores reflect the proportion of content that meets the threshold - not whether any single phrase does. A few standout sentences in otherwise generic content do not elevate the overall score.

How your score stays consistent

Most AI tools are non-deterministic - the same content fed in twice can produce different scores. TalentTell isn't. This section explains how, where the honest limits sit, and what to expect when something looks unexpected.

Same content, same score

Run TalentTell on the same content tomorrow, you'll get the same score - byte-for-byte identical to the score you got today. We hash your content; if we've already scored it, we don't ask the AI again. Same content, same score, every time.

Take Luxoft, one of the IT consultancies on our benchmark. We've scored Luxoft five times across our benchmark runs. Four returned Caregiver. One returned Sage. Same scoring engine, same scoring prompt - so what changed? The content did. On the Sage run, our scrape captured a substantially different set of visible job ads versus the other runs. Both reads are accurate to what was visible at scrape time. That's normal hiring behaviour: active companies cycle their visible roles, producing different but valid snapshots. When we re-scored the identical scrape from any single run, the result was byte-for-byte the same as the first time.

Same content, same score. Different content, different score. And the engine tells you when content changed.

When your content changes, we know

If your careers content changes between scores, the engine detects it. That's a feature, not a problem - drift signals are how the engine stays honest about what it's measuring.

On one of the companies in our benchmark with content-rich careers pages, drift detection surfaced unexpected variance across all four content categories - far more than rotation alone would explain. Investigation traced it to a timeout in our content-splitting layer that was reallocating content under specific conditions. We diagnosed the issue, fixed it, and re-ran. Subsequent runs confirmed stability. The point: drift detection didn't just flag noise. It pointed at a specific engineering bug we then fixed. That's the difference between unreliable AI and well-instrumented AI.

Scores have a noise floor on borderline cases

Even with byte-identical inputs and the cache active, the AI has small per-call variance on its initial judgement. We treat scores within ±5 of a category boundary as borderline cases - within the engine's noise floor.

A score of 47 vs 52 is meaningfully the same positioning. A score of 47 vs 58 is not - it crosses a threshold that affects the quadrant or how the result reads. Five points is the boundary between "essentially identical" and "different enough to matter." Within ±5, we don't claim precision we can't deliver.

Borderline cases treated honestly

Some companies sit genuinely between two patterns - their content carries traits of more than one archetype, or their scores land near a quadrant edge. The engine produces a single archetype call, but when a score sits within ±5 of a category boundary, that single call should be read qualitatively rather than treating it as definitive.

We see this most often when a brand sits genuinely between two patterns - a company that's part-Hero (achievement-led, ambitious) and part-Innocent (warm, inclusive), for example. Reading either as "the" definitive archetype would overstate what the content supports.

The right reading in those cases is qualitative: "tending toward Hero, with characteristics of Innocent" rather than "definitively Hero." The noise band tells you when the call is close enough to a boundary that the qualitative reading is the honest one.

Variance across your surfaces is the diagnostic

One last thing about variance. If your careers page tells one story and your job ads tell another, candidates notice that gap. The brand consistency score (covered later) measures it. The Talent Palette is built to close it. Variance across your own surfaces isn't a flaw in the measurement - it is the measurement.

How scoring works

Scores are proportion-based. If 80% of a company's content uses standard employer brand clichés and 20% contains genuinely distinctive language, the distinctiveness score reflects the 80%, not the 20%. The distinctive elements are noted as exceptions, but they do not inflate the overall score.

Headlines and headings carry more weight than buried body text, reflecting how candidates actually read careers pages - top-down, scanning for signal.

Specificity and distinctiveness are scored in separate calls so the first answer can't anchor the second; the other dimensions ride with the first call.

There is no forced distribution. We do not aim for a target percentage of companies in each quadrant. If 60% of companies genuinely produce generic content, 60% will land in the bottom-left. That is a finding, not a flaw in the methodology.

The positioning map

Every analysed company is plotted on a two-axis map:

• X-axis (Generic to Specific): The Specificity score. Content backed by concrete details, named programmes, and real numbers scores higher; vague, aspirational language scores lower.
• Y-axis (Cliched to Distinctive): The Distinctiveness score. Content that could not appear on a competitor's careers page scores higher; interchangeable, sector-standard language scores lower.

This creates four quadrants:

• The Communication Cloner (low X, low Y) - Generic and cliched. Vague claims delivered in language that could belong to any employer.
• The Hidden Gem (high X, low Y) - Specific but cliched. Real evidence and concrete details, but packaged in standard employer brand language.
• The Smooth Talker (low X, high Y) - Distinctive but generic. Original-sounding language that stands out, but without the specific evidence to back it up.
• The Standout (high X, high Y) - Specific and distinctive. Concrete evidence delivered in genuinely ownable language. The target.

Companies whose scores sit within ±5 of a quadrant boundary are borderline cases - the underlying content genuinely straddles two patterns. We treat those calls qualitatively rather than forcing a definitive quadrant. The "How your score stays consistent" section above explains how.

The 12 communication characters

Beyond the quadrant position, TalentTell identifies the dominant communication character - the archetype that best describes how an organisation talks about itself as an employer. These are based on established brand archetype theory, adapted specifically for employer brand communications.

The 12 characters are: Hero, Rebel, Explorer, Sage, Innocent, Jester, Ruler, Caregiver, Creator, Magician, Lover, and Regular Guy. Each has distinctive language patterns, motivations, and strengths. The character assignment is based on the overall pattern of language across all analysed content, not on any single phrase.

How benchmarking works

The comparison companies on the positioning map are real organisations whose public employer brand content has been analysed using the same methodology applied to your submissions.

How we collect content

For each company, we systematically crawl their career site and corporate pages to discover every publicly available employer brand page. We do not cherry-pick pages or editorially select which content to analyse. Every employer brand page we can access is included.

Job adverts are discovered automatically from the careers page. We follow links to individual job listings on known applicant tracking systems (Greenhouse, Lever, Ashby, Workday, and others). We scrape up to eight candidate listings, then deduplicate where ads share identical boilerplate (a common pattern - the same intro paragraph copied across multiple roles), and take the first five unique adverts that pass content quality checks. This gives us up to five genuinely distinct job-ad samples per company, selected by a fixed rule that applies to every company identically.

All content is collected automatically with no human editorial selection, ensuring every company is assessed on the same basis. We assess everything a company has published. No content is excluded or truncated.

The four content categories

We organise employer brand content into four categories, each weighted equally at 25%:

Within each category, all relevant pages are pooled together and scored as a whole. A company with seven pages of employer narrative content and a company with one page of employer narrative content are each scored on the quality of what they communicated - not on how many pages it took.

Why equal weighting?

Each content category contributes equally to the final score. This is a deliberate choice, not a default.

Candidates don't prioritise one touchpoint over another. They form impressions from everything they read - the careers page, the values statement, and the job ad that landed in their inbox. If your careers page is brilliant but your job descriptions are generic, candidates notice the gap. Equal weighting reflects the reality of how people experience your brand.

It eliminates the shop window problem. If one category counted for more, organisations could game their score by polishing that one surface while neglecting the rest. Equal weighting means you cannot fix your score by investing in one touchpoint. Consistency across all of them is what matters.

It lets the content speak for itself. A 20-word purpose statement gets the same weight as a 2,000-word careers page - and it should. Those 20 words represent months of deliberation. If they are genuinely distinctive and specific, they deserve that weight. If they are generic, that matters just as much.

Volume does not advantage anyone. Because our scoring measures proportions - what percentage of content is specific, what percentage is distinctive - publishing more pages does not improve your score. If anything, more content gives generic language more room to dilute strong messaging.

If a content category is not available for a company - for example, if no separate values or mission content exists - its weight is redistributed equally across the remaining categories.

Decoupled scoring

Specificity and distinctiveness are scored independently in separate analyses to prevent anchoring bias. The model assessing distinctiveness has zero knowledge of the specificity score, and vice versa. This ensures that a company's distinctiveness is measured on its own merits, not relative to how specific the content happens to be.

Company selection

Companies are selected globally across industries and size bands to provide meaningful comparison within your sector and beyond. Within each industry, we aim for representation across company sizes - from 50-person startups to 10,000+ enterprises. The final position on the map reflects the equally weighted average across all four content categories.

Minimum threshold

Companies must have employer narrative content - a careers page or equivalent - plus at least one additional content category to be included in the benchmark.

Companies below this threshold are excluded rather than scored unfairly. Scoring a company based on job ads and a leadership page without ever seeing their careers site would be like reviewing a restaurant based on the takeaway menu and the decor, but never tasting the food.

Exclusion is a fairness measure, not a limitation. Companies without a discoverable career site presence have not published enough for a fair assessment of their employer brand communication.

Brand consistency

Beyond the overall score, TalentTell measures the consistency gap between your career site and your job adverts. This is one of the most actionable insights in the benchmark.

A company with a strong, distinctive career site but generic job ads has a brand leakage problem. Candidates experience the weakest content at the moment of highest intent - when they are actually considering applying. The career site attracted them; the job ad failed to close the deal.

We compare your employer narrative scores against your job advertisement scores across all five dimensions. A small gap means your employer brand carries through consistently. A large gap means your recruitment marketing is not living up to what your career site promises - or, occasionally, that your job ads are stronger than your career site. Gaps below ±5 on any single dimension are within the engine's noise band and shouldn't be over-read on their own; what matters is the pattern across dimensions and the size of the largest gaps.

This is where the real advice lives. Your career site might be great, your values page might be great, but your job ads might be what is letting the side down. Or vice versa. Knowing where the gap is tells you exactly where to focus.

For every consistency-gap finding on a benchmark company, the audit data behind it captures which pages contributed to each side of the comparison and the per-category scoring underneath. If a finding looks wrong, we can pull the source pages and the per-category breakdown - the gap analysis is checkable against specific content, not asserted.

What this is (and what it isn't)

TalentTell is a directional insight tool, not a scientific measurement instrument. The 0-100 scores are AI-assisted structured assessments - informed judgements applied consistently through a clear framework, not calibrated measurements with statistical confidence intervals.

Think of it like a Michelin inspector. Nobody asks a Michelin inspector for inter-rater reliability coefficients or margin-of-error bars on their star ratings. They have a clear framework, they apply it consistently to every restaurant, and they explain their reasoning. The value comes from structured expert perspective applied at scale, not from pretending the result is physics.

TalentTell works the same way. AI applies a structured content analysis framework consistently to every company - the same dimensions, the same proportion-based scoring, the same cliché detection. What is defensible is this: AI-assisted content analysis using a structured framework applied consistently to every company. What is not defensible - and what we will never claim - is "scientifically validated scores" or "statistically significant positioning".

Same content, same score, every time. We don't rely on the AI being naturally deterministic - we cache by content hash. Run your analysis tomorrow on the same content and you'll get the same score, byte-for-byte. The "How your score stays consistent" section above explains why this matters and where the honest limits sit. Both user submissions and benchmark companies are analysed using the same AI model and the same scoring prompt, ensuring a fair comparison.

Scope - English-language content only

The benchmark is limited to English-language content. This is a deliberate limitation. Scoring employer brand communication is sensitive to idiom, tone, and nuance - qualities that don't survive translation reliably. Scoring a German company's German-language careers page using an English-calibrated framework would produce misleading results.

In practice, most global companies publish English-language versions of their careers content by default - it's their international hiring channel. When we encounter a company whose primary careers content is not in English, we exclude it from the benchmark rather than score it unfairly.

Transparency - show your working

The strongest defence of any methodology is transparency. For every benchmarked company on the positioning map, you can see exactly what we analysed and how we scored it:

• Pages analysed - every URL we crawled and included, organised by content category
• Per-category scores - how each content category scored across all five dimensions, with the per-page contributions underneath each score
• Brand consistency gap - the comparison between career site scores and job ad scores, with per-dimension breakdown
• Aggregated result - how the category scores combined into the final position on the map
• Audit row per scored category - the exact prompt the engine sent the AI, the exact response received, and metadata identifying which version of our scoring approach produced it

If a score looks wrong, we can show you the exact content the engine scored and the framework that was applied. The methodology is public; the per-page breakdown backs every score. The verdict is the score against our published framework, not the AI's prose commentary on it. This is a tool for informed conversation, not an oracle delivering verdicts.

Every benchmark score is backed by an immutable audit row, which we can produce on request. Challenges aren't "take our word for it" - they're a record we can pull.

How to read your score

"Why is my score low?"

Scores reflect proportions. If 80% of your content uses standard employer brand language - "passionate team", "collaborative environment", "we value diversity" - and 20% contains genuinely distinctive material, your score reflects the 80%. The distinctive elements are noted, but they do not define the overall score. Exceptions do not change the proportion.

"You missed some of our content"

We crawl career sites systematically and include every employer brand page we can access. If you believe we missed publicly available content, we are happy to re-run the analysis with updated URLs. The methodology is designed to capture everything - but websites change, pages move, and some content sits behind authentication or JavaScript rendering that our crawlers cannot access.

"Our career site is strong but our score is low"

Check the per-category breakdown. A strong career site with generic job ads will pull the overall score down - and that is intentional. Your job ads are the content candidates see at the moment of highest intent. If they do not reflect the same quality as your career site, that is a gap worth knowing about. The brand consistency indicator shows exactly where the disconnect is.

"We only have a simple careers page"

A company with one excellent, distinctive careers page can score higher than a company with twenty pages of generic content. Volume does not advantage anyone in this methodology. What matters is the proportion of your content that is specific, distinctive, and authentically written - whether that content lives on one page or fifty.

"What if I score the same content again - will I get the same result?"

Same content, same score, byte-for-byte. We hash your content and cache the result. If the content hasn't changed, the score is identical. If the content has changed - even slightly - the engine detects that and re-scores. Your report shows the most recent measurement; the engine retains an audit row capturing the prompt, the response, and the metadata behind it. For benchmark scores, we can produce that audit row on request.

"What if my archetype is different the next time you score me?"

If your content has changed, that's an honest reflection of what's now visible to a candidate reading your careers content - job rotation, hero copy updates, new testimonials all produce different but valid snapshots. If your content is identical and the archetype changes, that would indicate an engine problem; we have a cache invariant that prevents it and verification runs that confirm it consistently. Show us; we'd want to investigate too.

"What does drift mean on my measurement?"

Drift is our term for "your content changed since the last time we scored you." The engine flags it internally and captures the detail in an audit row. It's a signal, not a problem - it tells us whether your most recent measurement is comparable to the previous one. For benchmark scores, the audit row is queryable on request.

"My score is right on a boundary - which quadrant am I really in?"

Both, genuinely. Scores within ±5 of a quadrant boundary are borderline cases - the content sits between two patterns rather than landing definitively in one. The engine produces a single quadrant call, but when a score is within ±5 of a boundary, the right reading is qualitative ("tending toward Hidden Gem with characteristics of Smooth Talker") rather than treating the single call as definitive. The "How your score stays consistent" section earlier explains why this is honest rather than evasive.

Limitations

• Communication, not culture. TalentTell analyses how effectively you communicate your employer brand. It does not measure whether your actual culture is good, bad, or somewhere in between. A company could have a brilliant culture but communicate it poorly - and vice versa.
• Directional, not definitive. Scores are structured assessments, not scientific measurements. They indicate where your communication sits relative to others, not an absolute truth. Use them to start conversations and identify opportunities, not to declare winners and losers.
• Scrape determinism vs scoring determinism. Scoring is byte-deterministic on identical content. The scrape itself isn't always identical - some careers pages serve rotating widgets, geo-personalised copy, or A/B-tested hero text, so two scrapes on different days can capture slightly different bytes. Our drift detection surfaces this as a signal rather than masking it. Making the scrape itself more deterministic is on our roadmap.
• Point in time. Analysis reflects the content as it existed when it was submitted or scraped. Companies update their content regularly, and a score today may not reflect changes made tomorrow.
• Public content only. Benchmark analysis is based on publicly accessible careers pages and job adverts. Internal EVP documents, onboarding materials, and other non-public content are not included in benchmark scores (though you can submit them for your own analysis).
• AI-assisted, not AI-decided. The analysis is performed by AI applying a consistent framework. Like any analytical tool - human or automated - it is one input to inform decisions, not the final word. The transparency features exist precisely so you can evaluate the reasoning yourself.

Questions about our methodology? Get in touch at talenttellcontact@revealthereal.co.uk