Publish: Mistral AI Data Retention Policy: What You Need to Know

[harshikaalagh-netizen]

Article Ready for Publication

Title: Mistral AI Data Retention Policy: What You Need to Know
Author: Harshika
Date: 2026-03-17
Category: Guides

Branch: blog/mistral-data-retention-policy-1774865276889
File: apps/web/content/articles/mistral-data-retention-policy.mdx

---

Auto-generated PR from admin panel.

[netlify]

✅ Deploy Preview for hyprnote ready!

Name	Link
🔨 Latest commit	674c742
🔍 Latest deploy log	https://app.netlify.com/projects/hyprnote/deploys/69d60b20dcc1a20008ab20d6
😎 Deploy Preview	https://deploy-preview-4820--hyprnote.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify project configuration.

[netlify]

✅ Deploy Preview for char-cli-web canceled.

Name	Link
🔨 Latest commit	674c742
🔍 Latest deploy log	https://app.netlify.com/projects/char-cli-web/deploys/69d60b207d805b00085a7fe9

[github-actions]

Grammar Check Results

Reviewed 1 article.

Here's How Mistral Retains Your Data

📄 apps/web/content/articles/mistral-data-retention-policy.mdx

The article is well-written and professionally structured with clear explanations of Mistral's data governance policies. Only one grammatical issue was identified—a minor preposition error in a section heading. The content demonstrates good organization, consistent tone, proper punctuation placement with quotations, and no em dashes. Overall quality is high.

Found 1 issue:

📝 Grammar

Line 57

Use Mistral API to Take Meeting Notes Through Char

Grammar: 'Through' is incorrect preposition here; 'with' is the appropriate choice when indicating the tool or method being used.

📋 Suggested fix (click to expand)

## Use Mistral API to Take Meeting Notes With Char

---

Powered by Claude Haiku 4.5

---

AI Slop Check Results

Reviewed 1 article for AI writing patterns.

Here's How Mistral Retains Your Data

apps/web/content/articles/mistral-data-retention-policy.mdx

Score: 29/50 (NEEDS REVISION)

Dimension	Score
Directness	6/10
Rhythm	6/10
Trust	6/10
Authenticity	6/10
Density	5/10

This post reads as AI-generated because of recurring structural patterns, not vocabulary. The dominant issues are: (1) Binary antithesis in comparative sections (With X...With Mistral...), (2) Anaphoric repetition and double-beats that restate points for emphasis (particularly lines 34, 46, 52), (3) Conversational announcements that preview content instead of delivering it (lines 6, 12), (4) Clickbait heading formulas ('Is X Applicable?', 'How X Compares', imperative 'Use X to do Y'), and (5) Marketing framing that creeps in at the end (lines 54, 56). The technical content itself is accurate and well-organized, which is why the AI patterns stand out—they interrupt otherwise direct prose. The most egregious offenders are the heading templates and the opening comparative section (line 44), which uses a reversed setup pattern that LLMs deploy routinely. A human technical writer would state EU residency is the baseline and drop the With/With contrast. The post scores below 35/50 because of metronomic rhythm, conversational filler, and marketing language that violates the stated assumption that readers are competent enough to draw their own conclusions. The content itself is valuable; the presentation needs revision to remove the LLM rhetorical scaffolding.

Found 17 issues (0 high, 8 medium, 9 low)

MEDIUM — Likely AI Pattern

Line 11 — filler-phrase

That is not a minor detail.

Significance inflation crutch. The statement announces importance rather than demonstrating it. The subsequent sentences already make the case without this throat-clearing preface.

Suggested rewrite

Delete this sentence. The next sentence explains why it matters.

Line 15 — conversational-announcement

Here is how the policy works in practice.

Conversational announcement / preview. Tells the reader what's coming instead of showing it. The section headings that follow make this sentence redundant.

Suggested rewrite

Delete entirely. Move directly to the section heading.

Line 33 — clickbait-heading

Is Zero Data Retention Applicable?

Clickbait heading formula. 'Is X applicable?' is a marketing question that treats the reader as passive ('does this apply to me?') rather than descriptive ('what this section covers'). A technical heading should state what's being explained, not pose a question seeking confirmation.

Suggested rewrite

## Zero Data Retention

Line 37 — anaphoric-repetition

One important limitation: Zero Data Retention is not available on Le Chat. Because the consumer product relies on stored conversation history to function, ZDR cannot be applied there.

Anaphoric negation + significance inflation. 'One important limitation' is filler that inflates the point. The sentence structure (negation followed by causal explanation) is a standard AI rhythm. Collapse to a single direct statement.

Suggested rewrite

Zero Data Retention is unavailable on Le Chat because the product requires stored conversation history to operate.

Line 53 — antithesis-binary

The structural difference between Mistral and the US-based providers is data residency by default. With OpenAI, Anthropic, or Google, you are opting into EU data residency as an enterprise feature. With Mistral, EU residency is the baseline and US routing is the opt-in.

Binary antithesis pattern (With X... With Mistral...). The repetition of 'With' at sentence starts (anaphoric) builds artificial rhythm. The sentence structure—establishing one mode, then reversing it—is a textbook AI rhetorical move. Rewrite to state facts directly without the comparison setup.

Suggested rewrite

Mistral stores data in the EU by default. OpenAI, Anthropic, and Google require you to opt into EU residency. The burden of compliance is inverted: with Mistral, EU hosting is standard; with US providers, you must request it.

Line 55 — anaphoric-repetition

For organizations in the EU or those operating under GDPR, that reversal is meaningful. It shifts the burden of compliance from your team to Mistral's default configuration.

Significance inflation + repetition. 'That reversal is meaningful' announces importance; 'It shifts the burden' restates the same idea. The double-beat is metronomic and unnecessary. Collapse to a single claim.

Suggested rewrite

For EU organizations, this inversion means compliance is Mistral's responsibility, not yours.

Line 57 — clickbait-heading

Use Mistral API to Take Meeting Notes Through Char

Imperative heading formula ('Use X to do Y'). This is marketing copy dressed as a section label. It commands action rather than describing what the section covers. A technical heading should state the relationship, not push a product.

Suggested rewrite

## Mistral API with Char

Line 65 — marketing-framing

Download Char for macOS and use the AI provider your security team actually trusts.

The phrase 'the AI provider your security team actually trusts' is marketing positioning that implies competing providers are untrustworthy. The word 'actually' is a weasel word that sharpens the dig. A human writer would either remove this line or simply link to the download without the pitch.

Suggested rewrite

[Download Char for macOS](https://char.com/download/apple-silicon).

LOW — Subtle but Suspicious

Line 21 — conversational-announcement

You can do this by opening the Privacy menu in the Admin Console and disabling the toggle under Anonymous improvement data.

The phrasing 'You can do this' is a conversational softener. The sentence structure is also awkwardly procedural (five clauses chained with prepositions). A human writer would be more direct.

Suggested rewrite

Describe the same action with less procedural hand-holding: 'Disable the Anonymous improvement data toggle in Privacy settings.'

Line 39 — marketing-framing

Training Opt-Out in Plain Terms

The phrase 'In Plain Terms' is a marketing hedge that assumes the reader needs simplification. It's also a subtle form of clickbait framing. A descriptive heading ('How X Works') respects the reader's intelligence without announcing your intent to simplify.

Suggested rewrite

## How Training Opt-Out Works

Line 43 — anaphoric-repetition

If you are a paid API customer, you do not need to take any action to keep your data out of training pipelines. It is already excluded.

Repetition + binary antithesis disguised as reassurance. 'You do not need to take action' + 'It is already excluded' restates the same point. The double-beat feels like AI rhythm-building for emphasis. Combine into one sentence.

Suggested rewrite

Paid API customers don't need to opt out; your data is excluded by default.

Line 47 — marketing-framing

As an EU company, Mistral operates under GDPR natively rather than as a compliance overlay.

The phrase 'rather than as a compliance overlay' is unnecessarily wordy and imports jargon ('compliance overlay') that a human writer would avoid. The comparison to 'overlay' is marketing-speak.

Suggested rewrite

Mistral operates under GDPR natively, not as a compliance layer.

Line 49 — marketing-framing

Mistral does not currently publish a HIPAA Business Associate Agreement. For US healthcare organizations that need HIPAA coverage, Mistral is not the right choice without confirming a BAA is available for your specific plan.

The phrase 'Mistral is not the right choice' is marketing framing that positions Mistral as unfit rather than stating a capability gap. The follow-up 'without confirming a BAA is available' softens the statement unnecessarily. A human writer would be more direct about the limitation.

Suggested rewrite

Mistral does not publish a HIPAA Business Associate Agreement, so it cannot serve US healthcare organizations without custom negotiation.

Line 51 — clickbait-heading

How Mistral Compares With Other AI Providers

The heading 'How X Compares' is a marketing template used in product comparison articles. It's a listicle-style announcement rather than a descriptive label. A more direct heading names the comparison explicitly.

Suggested rewrite

## Mistral vs. US-Based Providers

Line 59 — staccato-fragments

Char is an open-source AI notepad for meetings that supports Mistral as one of its AI provider options. When you bring your own Mistral API key, your meeting data is handled under Mistral's API policy: 30-day retention for abuse monitoring, not used for training, stored in the EU by default.

Wordy setup + staccato final clause. 'When you bring your own Mistral API key' is conversational hand-holding. The final clause ('30-day retention, not used for training, stored in the EU') is a staccato list that fragments a complete thought. Simplify and flow into complete sentences.

Suggested rewrite

Char is an open-source meeting notepad that integrates Mistral as an AI provider. When you connect your Mistral API key, your meeting data follows Mistral's API policy: 30-day retention, not used for training, EU-stored by default.

Line 61 — anaphoric-repetition

If you activate ZDR on your Mistral API account, that protection applies to requests made through Char as well. Nothing is stored beyond the time needed to process the result.

Binary/conditional setup + redundant restatement. 'If you activate...that protection applies...Nothing is stored' restates the same point twice. The second sentence adds no new information. Combine or delete.

Suggested rewrite

If you enable ZDR on your Mistral API account, the same protection applies to requests through Char. No data is stored after processing.

Line 63 — marketing-framing

And if your data governance requirements change, or a different provider gets approved by your security team, you can switch inside Char without changing how your notes are stored or structured. Your files stay on your device regardless.

Conversational filler ('And if...or...') + marketing reassurance. The opening conditional is unnecessary throat-clearing. 'Your files stay on your device regardless' is a reassurance statement that reads like product copy. Human writers would be more direct about the feature.

Suggested rewrite

You can switch providers in Char without changing how notes are stored. Files remain on your device.

---

Powered by Claude Haiku 4.5 with stop-slop rules

[devin-ai-integration]

Blog Post Review: Humanizer + Stop-Slop

File: apps/web/content/articles/mistral-data-retention-policy.mdx

---

Humanizer Check (24 AI writing patterns)

Score: 39/50 (PASS)

Dimension	Score
Naturalness	8/10
Specificity	9/10
Voice	7/10
Rhythm	7/10
Conciseness	8/10

Overall the article is well-written with strong specifics (30 days, 5 years, named settings paths) and mostly free of major AI tells. The main issues are subtle: slightly stiff phrasing, significance inflation, and an overly even rhythm.

HIGH

Line	Original	Pattern	Suggested Rewrite
11	"That is not a minor detail."	#1 Significance inflation + #22 Filler	Delete entirely, or replace with "This matters." The next sentence already explains why.
15	"Here is how the policy works in practice."	#22 Filler phrase / throat-clearing	Delete. The next heading does this work.
53	"The structural difference between Mistral and the US-based providers is data residency by default."	#8 Copula avoidance (awkward "is data residency")	"Unlike US-based providers, Mistral stores data in the EU by default."

MEDIUM

Line	Original	Pattern	Suggested Rewrite
13	"For teams evaluating AI providers on data governance grounds, that starting point matters."	#1 Significance language + #7 AI vocabulary ("matters")	"For teams comparing AI providers on data governance, EU headquarters means GDPR compliance by default, not as an add-on."
25	"Account metadata follows a separate schedule."	Stiff phrasing	"Account metadata has different retention periods." or "Mistral keeps account metadata on a separate schedule."
39	"Training Opt-Out in Plain Terms"	#22 Filler phrase ("in plain terms")	"Training opt-out" or "Training Opt-Out Defaults by Plan"
49	"Mistral does not currently publish a HIPAA Business Associate Agreement. For US healthcare organizations that need HIPAA coverage, Mistral is not the right choice without confirming a BAA is available for your specific plan."	#23 Excessive hedging	"Mistral doesn't publish a HIPAA BAA. Healthcare organizations should confirm BAA availability for their plan before using Mistral."
55	"For organizations in the EU or those operating under GDPR, that reversal is meaningful."	#1 Significance language ("meaningful") + #7 AI vocabulary	"For EU organizations or those under GDPR, this reverses the compliance burden." Merge with next sentence.

LOW

Line	Original	Pattern	Suggested Rewrite
16-17	Title Case headings throughout (e.g. "What Mistral Stores by Default")	#16 Title Case in Headings	Consider sentence case: "What Mistral stores by default"
35	"ZDR is available on the API."	Terse/unclear	"Mistral offers ZDR for API customers."
37	"One important limitation: Zero Data Retention is not available on Le Chat."	Colon structure feels formulaic	"Zero Data Retention isn't available on Le Chat."

Patterns NOT found (good): No em dash overuse (#13), no emojis (#17), no collaborative artifacts (#19), no knowledge-cutoff disclaimers (#20), no sycophantic tone (#21), no negative parallelisms (#9), no bold overuse (#14), no curly quotes (#18), no synonym cycling (#11), no false ranges (#12), no inline-header lists (#15). The article avoids the majority of the 24 patterns.

---

Stop-Slop Check (phrases, structures, rhythm)

Score: 41/50 (PASS)

Dimension	Score
Directness	8/10
Rhythm	9/10
Trust	9/10
Authenticity	8/10
Density	7/10

The writing is strong overall. The dominant weakness is passive voice throughout (data "is stored," users "are opted in," features "are available"). The article avoids em-dashes, dramatic fragmentation, business jargon, adverbs, and meta-commentary.

Passive Voice (Primary Issue)

Most significant recurring pattern. Converting to active voice with Mistral as the named actor would improve authenticity.

Line	Original	Suggested Fix
19	"Mistral retains your inputs and outputs for 30 rolling days after the request is processed."	"...after processing the request."
21	"your conversations may be used for model improvement"	"Mistral may use your conversations for model improvement"
23	"Paid plans are handled differently."	"Paid plans work differently."
29	"By default, all data is hosted within the European Union."	"Mistral hosts all data within the EU by default."
31	"There is a US API endpoint available if you need it."	"Mistral offers a US API endpoint if you need it."
35	"ZDR is available on the API."	"Mistral offers ZDR on the API."
41	"Free Le Chat users are opted in to training by default"	"Mistral includes free Le Chat users in training by default."
48	"A Data Processing Addendum is available for all business customers."	"Mistral provides a Data Processing Addendum for all business customers."
49	"GDPR rights...can be exercised by contacting Mistral's privacy team"	"Contact Mistral's privacy team to exercise GDPR rights."
59	"your meeting data is handled under Mistral's API policy"	"Mistral handles your meeting data under its API policy"
63	"a different provider gets approved by your security team"	"your security team approves a different provider"

Structural Issues

Line	Original	Category	Suggested Fix
11	"That is not a minor detail."	Emphasis crutch / Telling instead of showing	Delete. The next sentence demonstrates why it matters.
15	"Here is how the policy works in practice."	Throat-clearing opener	Delete. The heading does this work.
25	"Account metadata follows a separate schedule."	False agency (metadata doesn't "follow")	"Mistral keeps account metadata on a separate schedule."
47	"Mistral operates under GDPR natively rather than as a compliance overlay."	Mild binary contrast ("rather than")	"Mistral operates under GDPR by design, not as an added layer."

Rhythm Issues

Line	Original	Category	Suggested Fix
41	Three parallel opt-in/opt-out statements	Metronomic rhythm (three parallel structures)	Collapse to a list or vary construction
53	"With OpenAI, Anthropic, or Google...With Mistral..."	Three-item list + metronomic "With X" pattern	Drop to two providers or restructure: "Mistral defaults to EU data residency. OpenAI and Anthropic require you to opt in as an enterprise feature."

Positive Patterns (No Issues Found)

• No em-dashes
• No dramatic fragmentation
• No "Not X, but Y" binary contrasts
• No business jargon
• No adverbs or filler words
• No meta-commentary
• No rhetorical questions
• Concrete specifics throughout (30 days, 5 years, 1 year, named UI paths)
• Good sentence length variation overall

---

Summary

Check	Score	Status
Humanizer (24 patterns)	39/50	PASS
Stop-Slop (phrases/structures/rhythm)	41/50	PASS
Combined	80/100	PASS

Top 3 Priority Fixes:

1. Delete throat-clearing sentences (lines 11, 15) — "That is not a minor detail." and "Here is how the policy works in practice." add no information.
2. Convert passive to active voice — Name Mistral as the actor throughout. This is the single biggest AI tell in the article.
3. Break metronomic patterns (lines 41, 53) — Vary the parallel opt-in/opt-out list structure and the "With X...With Y" comparison.

The article is well above the revision threshold on both checks. Content is substantive, specific, and well-sourced. The fixes above are refinements, not rewrites.

[devin-ai-integration]

Blog Post Review: Humanizer + Stop-Slop

Humanizer Check (24 AI writing patterns)

Score: 38/50 (PASS)

Dimension	Score
Naturalness	7/10
Specificity	9/10
Voice	7/10
Rhythm	6/10
Conciseness	8/10

The post is clean technical writing that largely avoids obvious AI patterns. Specificity is strong (actual retention periods, specific UI paths, named products). The main tell isn't pattern abuse — it's the absence of a human perspective. No reactions, no acknowledgment of uncertainty. It reads like a well-researched report, not someone sharing what they learned.

HIGH

No high-severity issues found.

MEDIUM

Line 11 — Pattern #21: Sycophantic/Servile Tone

That is not a minor detail.

Artificial emphasis telling the reader what to think.

Suggested rewrite

Remove entirely — the following sentence already demonstrates why it matters.

Line 13 — Pattern #1: Undue Emphasis on Significance

For teams evaluating AI providers on data governance grounds, that starting point matters.

Telling readers what to conclude rather than showing evidence.

Suggested rewrite

Remove this sentence — the evidence speaks for itself.

Line 15 — Pattern #19: Collaborative Communication Artifact

Here is how the policy works in practice.

Chatbot-style transitional phrase.

Suggested rewrite

Delete entirely. The heading that follows already signals what's coming.

Line 55 — Pattern #1: Undue Emphasis on Significance

For organizations in the EU or those operating under GDPR, that reversal is meaningful.

Telling readers what to conclude instead of showing.

Suggested rewrite

For organizations in the EU or those operating under GDPR, that reversal shifts compliance burden from your team to Mistral's infrastructure.

(Then remove the following sentence which says the same thing.)

LOW

Line 11 — Pattern #7/#4: Overused AI Vocabulary / Promotional Language

was built from the ground up under stricter privacy regulation

"From the ground up" is a cliché that adds no information.

Suggested rewrite

was designed under stricter privacy regulation

Line 37 — Pattern #11: Elegant Variation (Synonym Cycling)

Because the consumer product relies on stored conversation history to function, ZDR cannot be applied there.

"Consumer product" is a mild synonym cycle when you've already named "Le Chat."

Suggested rewrite

Le Chat needs stored conversation history to work, so ZDR cannot be applied there.

Line 41 — Pattern #10: Rule of Three

Free Le Chat users are opted in to training by default and can opt out via Privacy settings. Paid Le Chat and paid API users are opted out by default. ZDR API customers have no training risk because their data is never retained to begin with.

Three parallel statements feel slightly assembled, and "to begin with" is filler.

Suggested rewrite

Free Le Chat users are opted in to training by default and can opt out via Privacy settings. Paid Le Chat and paid API users are opted out by default. ZDR API customers have no training risk—Mistral never retains their data.

Line 35 — Pattern #22: Filler Phrase

Mistral does not retain your inputs or outputs beyond what is needed to return the result.

"Beyond what is needed" is wordy.

Suggested rewrite

Mistral only retains your data long enough to return the result.

Line 43 — Pattern #22: Filler Phrase

If you are a paid API customer, you do not need to take any action to keep your data out of training pipelines. It is already excluded.

Slightly wordy.

Suggested rewrite

Paid API customers don't need to do anything. Your data is already excluded from training.

Line 61 — Pattern #22: Filler Phrase

Nothing is stored beyond the time needed to process the result.

"Beyond the time needed" is wordy.

Suggested rewrite

Nothing is stored after processing.

---

Stop-Slop Check (phrases, structures, rhythm)

Score: 38/50 (PASS)

Dimension	Score
Directness	7/10
Rhythm	8/10
Trust	9/10
Authenticity	8/10
Density	6/10

The post is solid on authenticity and trust but suffers from passive voice overload. The content is direct and specific, but the delivery is often indirect. The single throat-clearing opener ("Here is how...") is minor but should be cut. The binary contrast in the comparison section is the only structural cliche, and it's mild.

Banned Phrases

Line 15 — Throat-clearing opener

Here is how the policy works in practice.

Matches "Here's what/this/that" pattern from phrases.md.

Suggested fix

Delete entirely. The next heading already signals what's coming.

Line 11 — Emphasis crutch / Telling instead of showing

That is not a minor detail.

Announces significance rather than demonstrating it.

Suggested fix

Delete. The following facts carry the weight on their own.

Structural Cliches

Line 53 — Binary contrast

With OpenAI, Anthropic, or Google, you are opting into EU data residency as an enterprise feature. With Mistral, EU residency is the baseline and US routing is the opt-in.

Mechanical "With X... With Y..." contrast pattern.

Suggested fix

OpenAI, Anthropic, and Google treat EU residency as an enterprise feature. Mistral makes it the default.

Line 55 — Vague declarative

that reversal is meaningful.

Announces importance without naming the specific thing.

Suggested fix

That matters for EU teams: compliance responsibility shifts to Mistral's configuration rather than your team's choices.

Passive Voice (primary issue — 10+ instances)

The post relies heavily on passive constructions. Key examples:

Line	Passive	Active rewrite
29	"all data is hosted within the EU"	"Mistral hosts all data within the EU"
35	"ZDR is available on the API"	"The API supports ZDR"
37	"ZDR cannot be applied there"	"You cannot use ZDR with Le Chat"
41	"Free Le Chat users are opted in"	"Mistral trains on free Le Chat data by default"
41	"Paid users are opted out by default"	"Mistral excludes paid users from training"
47	"A DPA is available for all business customers"	"Mistral provides a DPA for business customers"
47	"GDPR rights... can be exercised"	"You can exercise GDPR rights"
59	"your meeting data is handled under"	"Mistral handles your meeting data under"
61	"that protection applies to requests"	"Mistral applies that protection to Char requests"
61	"Nothing is stored beyond..."	"Mistral stores nothing after processing"

Rhythm Patterns

No em-dashes found. No three-item dramatic lists. Good paragraph length variation. Sentence lengths are reasonably varied. Minor metronomic quality in the training opt-out section (line 41) with three parallel statements.

---

Powered by humanizer + stop-slop

[devin-ai-integration]

Blog Post Review: Humanizer + Stop-Slop

File: apps/web/content/articles/mistral-data-retention-policy.mdx

---

Humanizer Check (24 AI writing patterns)

Score: 36/50 (PASS)

Dimension	Score
Naturalness	7/10
Specificity	9/10
Voice	6/10
Rhythm	6/10
Conciseness	8/10

The article has strong specifics (30-day retention, 5-year metadata, named UI settings paths, product names) and avoids most major AI tells. The main weaknesses are excessive passive voice, uniform rhythm, and a neutral corporate voice that lacks perspective. No em dashes, no emojis, no bold overuse, no curly quotes, no synonym cycling, no false ranges, no sycophantic tone, no knowledge-cutoff disclaimers.

HIGH

Line	Original	Pattern	Suggested Rewrite
11	"That is not a minor detail."	#1 Significance inflation + #27 Persuasive authority trope	Delete entirely. The next sentence already demonstrates why it matters.
15	"Here is how the policy works in practice."	#28 Signposting / #23 Filler phrase	Delete. The heading that follows does this work.
53	"With OpenAI, Anthropic, or Google, you are opting into EU data residency as an enterprise feature. With Mistral, EU residency is the baseline and US routing is the opt-in."	#9 Negative parallelism (binary contrast)	"OpenAI, Anthropic, and Google treat EU data residency as an enterprise add-on. Mistral makes it the default."

MEDIUM

Line	Original	Pattern	Suggested Rewrite
11	"was built from the ground up"	#23 Filler phrase / cliche	"was built under stricter privacy regulation"
13	"that starting point matters."	#1 Significance language	Delete, or replace with specific consequence: "...that changes the compliance calculation."
25	"Account metadata follows a separate schedule. Name and identity data is kept for 5 years... Email and phone number are retained for 1 year... Technical data... is kept for 1 rolling year."	#13 Passive voice x3, metronomic rhythm	"Mistral keeps account metadata on separate schedules: name/identity for 5 years after termination; email/phone for 1 year; connection logs for 1 rolling year."
39	"## Training Opt-Out in Plain Terms"	#23 Filler phrase ("in Plain Terms")	"## Training Opt-Out Defaults"
55	"that reversal is meaningful"	#1 Significance inflation	Merge with next sentence: "For EU organizations, this reverses the compliance burden: Mistral's defaults handle it, not your team."
65	"use the AI provider your security team actually trusts."	#4 Promotional language	Remove the editorializing. Just: "Download Char for macOS."

LOW

Line	Original	Pattern	Suggested Rewrite
11	"natively subject to GDPR"	#7 AI vocabulary ("natively" used abstractly)	"subject to GDPR by default"
19	"It is not used to train Mistral's models"	#13 Passive voice	"Mistral does not use it to train models"
21	"your conversations may be used for model improvement"	#13 Passive voice	"Mistral may use your conversations for model improvement"
23	"Paid plans are handled differently."	#13 Passive voice	"Mistral handles paid plans differently." or "Paid plans work differently."
29	"all data is hosted within the European Union"	#13 Passive voice	"Mistral hosts all data within the EU"
29	"Mistral explicitly states"	#23 Filler phrase	"Mistral prioritizes..." (drop "explicitly states")
31	"There is a US API endpoint available if you need it."	#13 Passive construction	"Mistral offers a US API endpoint."
33	"## Is Zero Data Retention Applicable?"	Question-form heading	"## Zero Data Retention on the API"
35	"ZDR is available on the API."	#13 Passive voice	"Mistral offers ZDR on the API."
37	"ZDR cannot be applied there"	#13 Passive voice	"Mistral cannot apply ZDR there"
41	"Free Le Chat users are opted in to training by default"	#13 Passive voice	"Mistral opts free Le Chat users into training by default"
43	"It is already excluded."	#13 Passive voice fragment	"Mistral already excludes it."
47	"A Data Processing Addendum is available"	#13 Passive voice	"Mistral provides a DPA for business customers."
47	"GDPR rights...can be exercised by contacting"	#13 Passive voice	"Contact Mistral's privacy team to exercise GDPR rights."
59	"your meeting data is handled under"	#13 Passive voice	"Mistral handles your meeting data under its API policy"

Patterns NOT found (good): No em dash overuse (#14), no emojis (#18), no collaborative artifacts (#20), no knowledge-cutoff disclaimers (#21), no sycophantic tone (#22), no negative parallelisms (#9, except the one noted), no bold overuse (#15), no curly quotes (#19), no synonym cycling (#11), no false ranges (#12), no inline-header lists (#16).

---

Stop-Slop Check (phrases, structures, rhythm)

Score: 34/50 (NEEDS REVISION if below 35)

Dimension	Score
Directness	6/10
Rhythm	7/10
Trust	8/10
Authenticity	7/10
Density	6/10

The writing has strong informational content and avoids most AI cliches (no em-dashes, no dramatic fragmentation, no "not X but Y" structures, no business jargon, no adverbs). The dominant weakness is systematic passive voice (20+ instances) that hides Mistral as the actor, and a handful of throat-clearing/emphasis phrases.

Banned Phrases

Line	Original	Category	Suggested Fix
11	"That is not a minor detail."	Emphasis crutch / Telling instead of showing	Delete. The next sentence demonstrates why.
15	"Here is how the policy works in practice."	Throat-clearing opener (matches "Here's what/this/that" pattern)	Delete. The heading does this work.
29	"Mistral explicitly states"	Filler adverb ("explicitly")	"Mistral prioritizes..." — drop the ceremony.
37	"One important limitation:"	Throat-clearing / emphasis crutch	Start directly: "Zero Data Retention is not available on Le Chat."
65	"your security team actually trusts"	Telling instead of showing ("actually matters" pattern)	Remove editorial. Just link the download.

Structural Cliches

Line	Original	Category	Suggested Fix
53	"With OpenAI, Anthropic, or Google... With Mistral..."	Binary contrast ("With X... With Y...")	"OpenAI, Anthropic, and Google treat EU residency as an enterprise feature. Mistral makes it the default."
55	"that reversal is meaningful"	Vague declarative (announces importance without naming the specific thing)	Merge into: "That reversal shifts compliance responsibility from your team to Mistral's configuration."
25	"Account metadata follows a separate schedule."	False agency (metadata doesn't "follow")	"Mistral keeps account metadata on a separate schedule."
47	"operates under GDPR natively rather than as a compliance overlay"	Mild binary contrast ("rather than")	"operates under GDPR by design, not as an added layer."

Passive Voice (primary systemic issue)

The post has 20+ passive constructions. Key instances:

Line	Passive	Active Rewrite
19	"after the request is processed"	"after processing the request"
19	"It is not used to train"	"Mistral does not use it to train"
21	"conversations may be used"	"Mistral may use your conversations"
23	"Paid plans are handled differently"	"Paid plans work differently"
25	"Name and identity data is kept"	"Mistral keeps name and identity data"
29	"all data is hosted within the EU"	"Mistral hosts all data in the EU"
35	"ZDR is available on the API"	"Mistral offers ZDR on the API"
41	"users are opted in to training"	"Mistral includes free users in training"
43	"It is already excluded."	"Mistral already excludes it."
47	"A DPA is available"	"Mistral provides a DPA"
47	"GDPR rights...can be exercised by"	"Contact Mistral's privacy team to exercise"
59	"meeting data is handled under"	"Mistral handles meeting data under"
63	"a different provider gets approved by your security team"	"your security team approves a different provider"

Rhythm Patterns

Line	Issue	Suggested Fix
25-27	Four consecutive sentences with identical "X is kept/retained for Y" structure	Collapse into semicolon list or table
41	Three parallel opt-in/opt-out statements (metronomic)	Vary construction or collapse to list
53	"With X... With Y..." parallel pattern + three-item list (OpenAI, Anthropic, Google)	Drop to two providers or restructure

Positive Patterns (No Issues)

• No em-dashes anywhere
• No dramatic fragmentation
• No business jargon (no "navigate," "landscape," "lean into")
• No adverbs or filler words (-ly)
• No meta-commentary
• No rhetorical questions answered immediately
• Concrete specifics throughout (30 days, 5 years, 1 year, named UI paths)
• Good paragraph length variation

---

Summary

Check	Score	Status
Humanizer (24 patterns)	36/50	PASS
Stop-Slop (phrases/structures/rhythm)	34/50	NEEDS REVISION
Combined	70/100	NEEDS REVISION

Top 3 Priority Fixes:

1. Convert passive to active voice throughout — Name Mistral as the actor. This is the single biggest AI tell (20+ instances). Fixing this alone would push the stop-slop score above 35.
2. Delete throat-clearing sentences (lines 11, 15) — "That is not a minor detail." and "Here is how the policy works in practice." add no information.
3. Break binary contrast and metronomic patterns (lines 25, 41, 53) — Vary the parallel retention schedule list, opt-in/opt-out list, and "With X...With Y" comparison structure.

The article's content is substantive, specific, and well-sourced. The issues are delivery patterns, not content problems. Converting passive constructions to active voice and cutting the handful of throat-clearing phrases would bring this comfortably above threshold.

---

Powered by humanizer + stop-slop

[devin-ai-integration]

Blog Post Review: Humanizer + Stop-Slop

File: apps/web/content/articles/mistral-data-retention-policy.mdx

---

Humanizer Check (24 AI writing patterns)

Score: 33/50 (NEEDS REVISION)

Dimension	Score
Naturalness	6/10
Specificity	8/10
Voice	6/10
Rhythm	7/10
Conciseness	6/10

The article has strong specifics (30-day retention, 5-year metadata, named UI paths) but is weighed down by pervasive passive voice (10+ instances), filler phrases, and significance inflation. The technical content is accurate; the prose style flags as AI-generated.

HIGH

Line	Original	Pattern	Suggested Rewrite
11	"That is not a minor detail."	#1 Significance inflation + #23 Filler	Delete entirely. The next sentence already demonstrates why EU headquarters matters.
15	"Here is how the policy works in practice."	#20 Collaborative communication artifact / throat-clearing	Delete. The section headings that follow already do this work.
53	"The structural difference between Mistral and the US-based providers is data residency by default."	#8 Copula avoidance + #1 Significance inflation ("structural difference")	"Mistral differs from US-based providers by making EU data residency the default."

MEDIUM

Line	Original	Pattern	Suggested Rewrite
11	"built from the ground up"	#7 AI vocabulary (stock phrase)	"designed"
13	"that starting point matters"	#1 Significance language + #5 Vague attribution	"For teams comparing AI providers on data governance, EU headquarters means GDPR compliance by default."
19	"It is not used to train Mistral's models"	#13 Passive voice	"Mistral does not use this data to train models"
21	"You can do this by opening the Privacy menu..."	#23 Filler phrase ("You can do this by")	"Open the Privacy menu in the Admin Console and disable the Anonymous improvement data toggle."
23	"Paid plans are handled differently."	#13 Passive voice	"Mistral treats paid plans differently."
25	"Account metadata follows a separate schedule."	Stiff phrasing / false agency	"Mistral keeps account metadata on different retention periods."
25	"Name and identity data is kept for 5 years" / "Email and phone number are retained for 1 year" / "Technical data...is kept for 1 rolling year"	#13 Passive voice (3 instances)	Use "Mistral keeps..." / "Mistral retains..." consistently.
37	"One important limitation:"	#1 Significance inflation	Delete. Start directly with "Zero Data Retention is not available on Le Chat."
37	"Because the consumer product relies on stored conversation history to function"	#11 Synonym cycling ("consumer product" when Le Chat was just named)	"Le Chat needs stored conversation history to work, so ZDR cannot be applied."
43	"If you are a paid API customer, you do not need to take any action...It is already excluded."	#23 Filler phrase (entire conditional restates previous paragraph)	"Paid API customers are already excluded from training pipelines."
45	"Mistral does not currently publish"	#24 Excessive hedging ("currently")	"Mistral does not publish a HIPAA BAA."
47	"GDPR rights...can be exercised by contacting"	#13 Passive voice	"Contact Mistral's privacy team to exercise GDPR rights."
55	"that reversal is meaningful. It shifts the burden of compliance from your team to Mistral's default configuration."	#1 Significance inflation + restated double-beat	"For EU organizations, compliance responsibility shifts to Mistral's default configuration." (merge into one sentence)
65	"use the AI provider your security team actually trusts"	#4 Promotional language	"Download Char for macOS."

LOW

Line	Original	Pattern	Suggested Rewrite
17-57	Title Case headings throughout	#17 Title Case in Headings	Consider sentence case: "What Mistral stores by default"
31	"There is a US API endpoint available if you need it."	#23 Filler ("if you need it")	"Mistral also offers a US API endpoint."
35	"beyond what is needed to return the result"	#23 Filler (wordy)	"beyond the time required to return the response"
63	"Your files stay on your device regardless."	#4 Promotional language ("regardless" overstates)	"Your files stay on your device."

Patterns NOT found (good): No em dash overuse (#14), no emojis (#18), no curly quotes (#19), no knowledge-cutoff disclaimers (#21), no sycophantic tone (#22), no negative parallelisms (#9), no bold overuse (#15), no synonym cycling (#11 — one minor instance), no false ranges (#12), no inline-header lists (#16).

---

Stop-Slop Check (phrases, structures, rhythm)

Score: 34/50 (NEEDS REVISION)

Dimension	Score
Directness	7/10
Rhythm	6/10
Trust	8/10
Authenticity	6/10
Density	7/10

The article is strong on trust (concrete specifics, no hand-holding) and directness, but passive voice throughout and a few structural cliches drag it below the threshold. No adverbs, no business jargon, no em-dashes, no dramatic fragmentation.

Banned Phrases

Line	Original	Category	Suggested Fix
2	"Here's How Mistral Retains Your Data"	Throat-clearing opener ("Here's how")	"How Mistral Retains Your Data" or "Mistral's Data Retention Policy"
11	"That is not a minor detail."	Emphasis crutch / Telling instead of showing	Delete. The facts carry the weight.
15	"Here is how the policy works in practice."	Throat-clearing opener ("Here is how")	Delete entirely.
37	"One important limitation:"	Meta-commentary (announces structure)	Delete. Start with "Zero Data Retention is not available on Le Chat."

Structural Cliches

Line	Original	Category	Suggested Fix
47	"operates under GDPR natively rather than as a compliance overlay"	Binary contrast ("rather than")	"Mistral operates under GDPR natively, not as a compliance layer."
53	"With OpenAI, Anthropic, or Google...With Mistral..."	Binary antithesis (anaphoric "With X...With Y")	"OpenAI, Anthropic, and Google treat EU residency as an enterprise feature. Mistral makes it the default."
55	"that reversal is meaningful"	Vague declarative	Name the specific benefit: "compliance responsibility shifts to Mistral's configuration."
33	"Is Zero Data Retention Applicable?"	Clickbait heading formula	"Zero Data Retention"
51	"How Mistral Compares With Other AI Providers"	Marketing heading template	"Mistral vs. US-Based Providers"
57	"Use Mistral API to Take Meeting Notes Through Char"	Imperative heading (marketing)	"Mistral API with Char"

Passive Voice (Primary Issue — 11 instances)

Line	Passive	Active Rewrite
19	"the data is deleted"	"Mistral deletes the data"
19	"It is not used to train"	"Mistral does not use it to train"
23	"Paid plans are handled differently"	"Mistral handles paid plans differently"
29	"all data is hosted within the EU"	"Mistral hosts all data within the EU"
35	"ZDR is available on the API"	"Mistral offers ZDR on the API"
41	"Free Le Chat users are opted in"	"Mistral opts free Le Chat users in"
41	"Paid users are opted out by default"	"Mistral excludes paid users from training"
47	"A DPA is available for all business customers"	"Mistral provides a DPA for business customers"
47	"GDPR rights...can be exercised"	"You can exercise GDPR rights"
59	"your meeting data is handled under"	"Mistral handles your meeting data under"
61	"Nothing is stored beyond the time needed"	"Mistral stores nothing after processing"

Rhythm Issues

Line	Original	Category	Suggested Fix
41	Three parallel opt-in/opt-out statements	Metronomic rhythm (three parallel constructions)	Vary structure or collapse into a list
53	"With OpenAI, Anthropic, or Google"	Three-item list	Drop to two providers or restructure
43	"you do not need to take any action...It is already excluded."	Immediate restatement (double-beat)	Combine: "Paid API customers don't need to opt out; your data is excluded by default."

Positive Patterns (No Issues Found)

• No adverbs (-ly words)
• No business jargon
• No dramatic fragmentation
• No em-dashes
• No meta-commentary (beyond lines 11, 15, 37)
• No rhetorical questions
• Concrete specifics throughout (30 days, 5 years, 1 year, named UI paths)

---

Summary

Check	Score	Status
Humanizer (24 patterns)	33/50	NEEDS REVISION
Stop-Slop (phrases/structures/rhythm)	34/50	NEEDS REVISION
Combined	67/100	NEEDS REVISION

Top 3 Priority Fixes:

1. Convert passive to active voice — 11 instances. Name Mistral as the actor throughout. This is the single biggest AI tell.
2. Delete throat-clearing sentences (lines 2, 11, 15, 37) — "That is not a minor detail," "Here is how the policy works in practice," "One important limitation:" add no information.
3. Break structural cliches — Rewrite the "With X...With Y" comparison (line 53), fix clickbait heading formulas (lines 33, 51, 57), and collapse double-beat restatements (lines 43, 55).

The technical content is accurate and well-organized. The fixes above are about delivery, not substance.