sfeir.com
Evidence collected August 31, 2026 · 33 polite, robots.txt-respecting requests · every claim cites what it saw
D
54.6/100 — advocacy readiness(reduced confidence: parts of the evidence were unobservable — see below)
Graded because the reader classified enough of this site to trust the measurement. For scale: of the 350 companies in this corpus, only 126 earned a grade — for the rest I withheld it rather than grade a site the reader could not read.
Read this before the numbers
This tool was built in a few days, on a stack of stated assumptions. I didn't have the time — or the inside knowledge of what buyers in this category actually optimize for — to be certain these are the metrics that matter most. The four categories and their weights are editorial judgment, not measurement: nobody has proven that these things cause employees to post, and I don't claim otherwise.
Where the evidence comes from: three of the four categories read only what your company has published on its own public site — the part you can fix on Monday. Half of Employee & Culture comes from somewhere else: aggregate counts of how many of your people posted in the last 90 days, from a licensed third-party index of public professional-network activity. Nothing here was scraped from LinkedIn by this tool. Those counts are records in an index, not a census — they undercount, they undercount small companies most, and you can't check them yourself — so each one says so where it appears, a zero is reported as "none in the index" and never as "nobody posts," and a company with fewer than 25 people in the index gets the number reported but not scored. No individual is named anywhere in this report.
To prioritize speed, this particular report was loaded using cached data; it was generated from evidence originally collected on August 31, 2026 rather than live. Reports are cached for 30 days. Every observation below is dated and cited.
What you can hold me to: every number cites its evidence and says where it came from, nothing unseen is guessed at, robots.txt is respected on every fetch, and when this tool can't read a site it says so instead of issuing a grade.
What was found — most fixable first
Ranked by impact × ease of fixing, not by category. Counts describe the pages actually read, and each finding cites them.
Your homepage share card is missing og:image
The homepage lacks og:image. When anyone — including your own employees — shares your homepage link, the preview card renders incomplete (no image means most platforms show a bare grey box).
Evidence: homepage HTML.
The fix: Add the missing tag to the homepage template: og:image. Minutes of work, visible on every share afterward.
0 of 24 sampled pages carry a visible share control
None of the 24 content pages sampled carry a visible share control. Nothing on the page invites a reader — or an employee — to share it.
Evidence — content page sample
- https://www.sfeir.com/articles/4-piliers-ia-mesh-adaptatif/
- https://www.sfeir.com/articles/automatisation-processus-agents-conversationnels/
- https://www.sfeir.com/articles/claude-code-interne-sfeir-dogfooding/
- https://www.sfeir.com/articles/confidential-computing-tee-protection-donnees/
- https://www.sfeir.com/articles/context-engineering-multi-agents/
- https://www.sfeir.com/articles/deploiements-durcis-agents-nemoclaw-archestra/
- https://www.sfeir.com/articles/gestion-changements-grande-echelle-itil-augmente/
- https://www.sfeir.com/articles/google-cloud-sfeir-16-ans-partenariat/
- … and 16 more in the sample
The fix: Add share links or buttons to the content template. Plain links to the platform share endpoints work without any tracking script.
0 of 24 sampled pages name a human author
None of the 24 sampled content pages carry a named human author — no byline, no author schema, no rel=author. Content nobody signs is content nobody feels ownership of; advocacy runs on employees being visible in the work.
Evidence: content page sample — 0 of 24 sampled pages.
The fix: Add author bylines to content pages — a name, ideally with a link or an author block in the Article schema. Start with new posts; backfilling can wait.
Your job listings are invisible without JavaScript
Your careers page (https://www.sfeir.com/en/careers/jobs) returns HTML, but the job listings are injected client-side — a plain reader sees a page with no roles on it. Machine readers that do not execute JavaScript (many crawlers and AI assistants) cannot see that you are hiring.
Evidence — careers page probes
- https://www.sfeir.com/en/careers/jobs
- https://www.sfeir.com/en/careers
The fix: Server-render at least the role titles and locations, or publish jobs in structured data (schema.org JobPosting).
No culture or life-at pages anywhere in your sitemap
Of 1,214 canonical URLs, zero classify as culture content — no life-at pages, employee stories, or team spotlights. This is the most common gap this tool finds, and the most directly advocacy-shaped one: it is the content employees share proudest and first. Your people's own posting is scored separately, below.
Evidence: sitemap URL classification.
The fix: Start small: one employee-story post a month, indexed in the sitemap. The section can grow from there.
What's already working
All tracked AI user-agents may read your site
robots.txt permits all 25 tracked AI user-agents at the site root, so AI search engines and assistants that honor robots.txt can read and cite you.
Evidence: robots.txt.
What could not be seen
Anything the reader could not observe is listed here rather than guessed at or silently skipped.
Content freshness is not observable here
Your sitemap publishes lastmod dates, but they look machine-generated rather than edit history (29.6% of all lastmod values fall on a single day (2026-08-31), which is the signature of a site-wide rebuild rather than individual edits.; The entire inventory's lastmod range spans only 152 days, which is implausibly short for 1215 pages and suggests the dates were regenerated.), so I cannot read them as freshness and have not guessed.
Evidence: sitemap lastmod analysis.
The fix: Emit real last-modified dates in the sitemap. Credible freshness data helps every machine reader — search, AI, and tools like this one — represent you accurately.
For the record
llms.txt is present
An llms.txt file is served. Reported for completeness: Google has said it does not support llms.txt, no major AI vendor has confirmed consuming it, and large-scale studies found no correlation with AI citations. I record it and weight it at zero.
Evidence: llms.txt fetch.
How the numbers break down
Weights: Content Supply 30% · Shareability 25% · Employee & Culture 25% (half what the site says, half what employees do) · AI Discoverability 20%. The weights are editorial judgment about what an advocacy program draws on — not the result of a study. Every row cites the observation behind it and says whether it was read from your site or from the licensed index.
Content Supply — 43.2/100 (reduced confidence)
| Signal | Points | What was observed |
|---|---|---|
| Volume floor | 23.8/40 | 318 shareable pages after collapsing localized duplicates |
| Section diversity | 9/30 | 2 shareable section(s): blog, news_press |
| Shareable share | 10.4/30 | 26.2% of the canonical inventory is shareable content |
Recency was not scored: this site publishes lastmod dates, but they look machine-generated rather than edit history, so they cannot be read as freshness.
Shareability — 61.3/100
| Signal | Points | What was observed |
|---|---|---|
| Open Graph image — without one, a shared link renders as a bare grey box | 26.3/35 | 18 of 24 sampled pages (75%) |
| Open Graph title on the shared card | 15/15 | 24 of 24 sampled pages (100%) |
| Open Graph description on the shared card | 15/15 | 24 of 24 sampled pages (100%) |
| A visible share control on the page | 0/30 | 0 of 24 sampled pages (0%) |
| Twitter/X card tags (near-zero weight: X falls back to Open Graph) | 3/3 | 24 of 24 sampled pages (100%) |
| Canonical URL (near-zero weight: SEO hygiene, not a sharing condition) | 2/2 | 24 of 24 sampled pages (100%) |
Employee & Culture — 35.9/100
| Signal | Points | What was observed |
|---|---|---|
| What the site says (50 of 100) | ||
| Culture / life-at pages | 0/15 | 0 culture page(s) in the canonical inventory |
| Careers page reachable | 7.5/7.5 | A careers page was found and retrieved |
| Careers depth | 7.5/7.5 | 57 careers URL(s) in the sitemap |
| Named authors on content | 0/12.5 | 0% of sampled pages carry a named author |
| Employee-story signals | 0/7.5 | No culture URLs existed to sample, so no employee-story signal could be checked |
| What employees do (50 of 100) | ||
| Employees posting (last 90 days) | 13.9/25 | 84 of 763 indexed employees posted in the last 90 days (11%). licensed third-party index (not independently verifiable); 763 of 1066 employees covered. |
| Decision-makers posting | 6.1/15 | 13 decision-makers posted in the last 90 days (1.7% of indexed employees). licensed third-party index (not independently verifiable). |
| Company page posting cadence | 1/10 | 13 company post(s) in the last 90 days in the index. licensed third-party index (not independently verifiable). |
A qualitative read of the careers page (boilerplate vs. real human stories) is not included in this number.
Half of this category is measured from employee posting activity in a licensed third-party index (763 of 1066 employees covered). Those counts are records in a third-party index, a lower bound the visitor cannot verify independently; the other half is what the website itself shows.
AI Discoverability — 86.6/100
| Signal | Points | What was observed |
|---|---|---|
| AI crawler access | 45/45 | All 25 tracked AI user-agents are permitted at the site root |
| Content-Signal coherence | 9/15 | No Content-Signal declared. Neither a grant nor a restriction — scored neutrally. |
| Structured data | 32.6/40 | 7 schema.org type(s) on the homepage; 79.2% of sampled content pages carry Article schema |
| llms.txt | 0/0 | Present. Scored at zero weight deliberately: Google has said it does not support llms.txt, no major AI vendor has confirmed consuming it, and large-scale studies found no correlation with AI citations. |
robots.txt compliance is voluntary and advisory. This reports the declared policy; it is not a claim that any crawler was actually blocked.