✅ Deploy Preview for char-cli-web canceled.

✅ Deploy Preview for hyprnote ready!

Name	Link
🔨 Latest commit	21082e5
🔍 Latest deploy log	https://app.netlify.com/projects/hyprnote/deploys/69ca4b2b4ff89d0008d11ba4
😎 Deploy Preview	https://deploy-preview-4818--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.

Grammar Check Results

Reviewed 1 article.

Everything You Should Know About Google Gemini Data Retention Policy

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

The article is well-written and comprehensive with clear structure and logical flow. However, there are several style rule violations regarding em dashes that need attention per the specified guidelines. One title capitalization issue was also identified. The punctuation, grammar, and clarity are otherwise strong throughout. Recommend addressing the em dash replacements and the title capitalization to bring the piece fully into compliance with the stated style rules.

Found 6 issues:

🔸 Em Dashes

Line 11

That changes the risk profile entirely.

This sentence introduces a contrast that would benefit from an em dash, but per style rules, em dashes should be flagged. Consider: 'That changes the risk profile entirely - or rewrite to avoid the em dash.

That changes the risk profile entirely—

Line 34

Deleting your Gemini activity does not remove those reviewed conversations.

Em dash usage detected. Per style rules, replace with regular dash or rewrite: 'Deleting your Gemini activity does not remove those reviewed conversations -'

Deleting your Gemini activity does not remove those reviewed conversations—

Line 44

The case Thele v. Google LLC, filed in the Northern District of California, alleges a violation of the California Invasion of Privacy Act (CIPA).

Em dashes used for parenthetical clause. Replace with regular dashes per style rules, or remove the clause and rewrite.

The case Thele v. Google LLC - filed in the Northern District of California - alleges a violation of the California Invasion of Privacy Act (CIPA).

Line 72

This cached data has its own TTL (Time-to-Live) setting that you control, and it sits outside the standard 55-day abuse monitoring window. It is a flexible feature, but it means you may have data on Google's servers for longer than the default if you use it.

Two independent clauses separated by a period could be connected with an em dash or regular dash for better flow. Consider using a regular dash per style rules.

This cached data has its own TTL (Time-to-Live) setting that you control, and it sits outside the standard 55-day abuse monitoring window - it is a flexible feature, but it means you may have data on Google's servers for longer than the default if you use it.

🔹 Punctuation Placement

Line 60

Google's Gemini Apps Privacy Hub states that trained human reviewers check conversations to evaluate whether responses are low quality, inaccurate, or harmful.

Per British style punctuation rules, there should be a space before the URL in markdown link syntax for clarity.

Google's [Gemini Apps Privacy Hub] (https://support.google.com/gemini/answer/13594961) states that trained human reviewers check conversations to evaluate whether responses are low quality, inaccurate, or harmful.

📝 Grammar

Line 96

Use Gemini's API for Meeting Notes Through Char

'Through' should not be capitalized in a title unless it is the first word. Follow standard title capitalization rules.

Use Gemini's API for Meeting Notes through Char

Powered by Claude Haiku 4.5

AI Slop Check Results

Reviewed 1 article for AI writing patterns.

Everything You Should Know About Google Gemini Data Retention Policy

apps/web/content/articles/google-gemini-data-retention-policy.mdx

Score: 25/50 (NEEDS REVISION)

This post exhibits strong AI writing patterns, particularly in structural construction rather than vocabulary. The dominant issues are: (1) Antithesis / binary framing throughout ('X isn't Y—it's Z', 'When you evaluate OpenAI...When you evaluate Google'), (2) Metronomic rhythm in section-level structures (HIPAA/GDPR sections mirror each other; Comparison section uses four parallel sentences), (3) Conversational announcements that preview content ('Here is the full picture', 'Here is the part that catches'), (4) Clickbait heading formulas ('How X Works?', 'The X Story', imperative commands like 'Use Gemini's API'), (5) Staccato list structures for artificial emphasis (three-sentence rules in Workspace section), and (6) Marketing framing throughout the Char section (testimonial language about control, agency, and trust). The table and most technical details are solid, but the connective prose—especially section transitions, openings, and closings—reads like an LLM arguing a point rather than a technical writer documenting facts. The final three paragraphs are particularly weak, using binary contrasts and emotional appeals instead of direct comparison. Score below 35/50: needs significant structural revision to sound human.

Found 33 issues (1 high, 17 medium, 15 low)

HIGH — Obvious AI Tell

Line 104 — antithesis-binary

That is the practical difference between using a consumer AI product and choosing your own stack. The first gives you a privacy policy. The second gives you control.

Antithesis setup with metronomic rhythm: 'That is the practical difference. The first X. The second Y.' is a classic three-part binary contrast structure. 'practical difference' is weak framing. The two-sentence list creates artificial emphasis. This should be collapsed into one direct comparison.

The difference: a consumer product gives you a privacy policy; your own stack gives you control.

MEDIUM — Likely AI Pattern

Line 11 — antithesis-binary

Every AI provider in this series stores your conversations to some degree. Google Gemini does that too. But Gemini sits inside an ecosystem that already holds your email, your calendar, your location history, and your search activity. That changes the risk profile entirely.

Antithesis pattern: 'X does that. But Y changes it entirely.' Sets up negation (what others do) before the claim. The four-item list ('email, your calendar, your location history, and your search activity') is metronomic enumeration typical of AI writing.

Google Gemini stores conversations like other AI providers, but within an ecosystem that already holds your email, calendar, location history, and search activity. This fundamentally changes the privacy risk.

Line 13 — anaphoric-repetition

When you evaluate OpenAI or Anthropic, you're asking what they do with your AI conversations. When you evaluate Google, you're asking what they do with your AI conversations and everything else they already know about you.

Anaphoric repetition with metronomic rhythm: 'When you evaluate X...When you evaluate Y...' followed by parallel clause structure creates the AI rhetorical cadence. The repetition of 'what they do with your AI conversations' is a classic LLM device to build false emphasis.

Google's risk profile differs from OpenAI or Anthropic because it combines AI conversation data with email, location, search history, and calendar data already in the ecosystem.

Line 34 — conversational-announcement

Here is the part that catches most people off guard: if Google's human reviewers look at one of your conversations, that conversation is retained for up to 3 years, disconnected from your Google Account. Deleting your Gemini activity does not remove those reviewed conversations. They persist regardless.

Conversational announcement ('Here is the part that catches...') previews the claim instead of stating it directly. 'They persist regardless' is repetitive given the prior sentence. The phrasing reads like building dramatic tension rather than delivering information.

If Google reviewers examine a conversation, it's retained for 3 years disconnected from your account—even after you delete your Gemini activity. Deleting the original doesn't remove the reviewed copy.

Line 36 — antithesis-binary

Google says it tries to anonymize or disconnect reviewed chats before humans see them. But if you typed personal information directly into the conversation, that information is part of the message and visible to the reviewer.

Binary antithesis structure: 'Google says it tries X. But if you Y, then Z.' The 'tries to' + 'but' pattern sets up a negation before the actual risk. This is a textbook AI rhetorical move.

Google attempts to anonymize reviewed conversations before human review, but personal information you typed remains visible to the reviewer.

Line 46 — antithesis-binary

Google's position is that this data is used only to power features for individual users, not to train its public AI models. That is a meaningful distinction. But for anyone evaluating Gemini as a provider for sensitive work, the default-on access to your full email history is worth knowing about upfront.

Marketing framing and antithesis: 'Google's position is X. That is a meaningful distinction. But Y is worth knowing.' The phrase 'worth knowing about upfront' reads like testimonial framing (positioning the reader benefit). Combining the affirmation with the contradicting 'But' creates artificial balance.

Google states this data powers individual features, not model training—a meaningful distinction. For anyone handling sensitive work, the default-on Gmail access is worth knowing about upfront.

Line 66 — antithesis-binary

Google advises users not to enter confidential information they would not want a reviewer to see. That is practical advice, but it places the compliance burden squarely on users rather than on the product design.

Antithesis with significance inflation: 'That is practical advice, but...' creates the setup-and-counter rhetorical move. 'places the compliance burden squarely on users rather than on the product design' uses the word 'squarely' (intensifier) and frames responsibility as a burden (anthropomorphizing responsibility). The antithesis structure artificially builds tension.

Google advises users to avoid entering confidential information. This shifts compliance responsibility from product design to user behavior.

Line 74 — metronomic-rhythm

The second is Session Resumption for the Gemini Live API. If you use this to keep a voice session active across interruptions, Google caches that session data in memory for up to 24 hours. That is a separate and shorter window from the 55-day abuse logs, but it is worth knowing if you are using the voice API for sensitive conversations.

Metronomic parallel structure: 'The second is...' mirrors 'The first is...' (list formula). 'That is a separate and shorter window from..., but it is worth knowing' uses antithesis ('shorter', 'but') and the phrase 'worth knowing' (marketing framing). The phrase 'if you are using the voice API for sensitive conversations' is audience hedging rather than direct advice.

Session Resumption for Gemini Live caches voice data in memory for 24 hours—a shorter window than the 55-day abuse logs, but relevant for sensitive conversations.

Line 82 — staccato-fragments

For Google Workspace Enterprise customers, the rules are more protective. Prompt content is not used to train Google's public AI models without explicit permission. Human review of your conversations does not happen without your organization's consent. Workspace admins can shorten or fully disable prompt storage for their domain.

Staccato three-sentence structure with metronomic rhythm: each sentence follows the pattern 'Rule X. Rule Y. Rule Z.' The passive voice ('is not used', 'does not happen') repeats the negation structure. Combining into one sentence with active voice ('customers get X, Y, and Z') is more direct and human.

Google Workspace Enterprise customers get stronger protections: no model training without permission, no human review without consent, and admins can disable or shorten storage.

Line 84 — metronomic-rhythm

On HIPAA: as of September 30, 2025, Gemini for Workspace is included under Google's HIPAA Business Associate Addendum. As of Q1 2026, this is fully stable and has been widely adopted by healthcare organizations. You need a signed BAA and the appropriate project flags enabled through the Admin Console. The free consumer version is not covered and should not be used with protected health information.

Metronomic list structure with parallel openers ('On HIPAA', 'As of', 'You need'...) and excessive date specificity ('September 30, 2025', 'Q1 2026'). The phrase 'this is fully stable and has been widely adopted' is filler intensifier language. Four sentences could be consolidated into two. The negation at the end ('is not covered and should not be used') is a secondary point that buries the actual requirement.

HIPAA: Gemini for Workspace is covered under Google's Business Associate Addendum as of September 2025 (stable as of Q1 2026). Requires a signed BAA and Admin Console flags. The free consumer version is not HIPAA-covered.

Line 86 — metronomic-rhythm

On GDPR: Google offers a Data Processing Addendum for Workspace customers. Consumer accounts are subject to Google's standard privacy policy. EU consumer users have GDPR rights including access and deletion, but those rights do not remove reviewed conversations from Google's systems during the 3-year retention window.

Metronomic structure mirrors HIPAA section (heading + list). The phrase 'but those rights do not remove' is antithesis ('have rights, but they don't apply to X'). Three sentences repeat 'Google' and 'consumers' in parallel positions. Passive voice ('are subject to') and filler ('including') pad the text.

GDPR: Workspace customers get a Data Processing Addendum. EU consumers have access and deletion rights, but these don't remove reviewed conversations from Google's 3-year retention window.

Line 90 — anthropomorphization

OpenAI and Anthropic each had a single data retention policy story to tell. Google's story is more layered because of how deeply Gemini integrates with the rest of their products.

Anthropomorphization and marketing framing: 'had a story to tell' personifies companies and uses narrative metaphor. 'Google's story is more layered because of how deeply Gemini integrates' is metaphorical and vague. A human technical writer would compare the actual policy structures, not frame them as narratives.

OpenAI and Anthropic have one data retention policy each. Google's is layered across multiple products and integrations.

Line 92 — metronomic-rhythm

The consumer product connects to Gmail, Calendar, Drive, and location data. The retention periods are longer by default than either OpenAI or Anthropic. The human review window is the longest of the three. And in the US, the Gmail integration was enabled without asking users first.

Metronomic four-sentence structure with parallel openers ('The consumer product...', 'The retention periods...', 'The human review window...', 'And in the US...'). Each sentence follows the same length and cadence. The enumeration 'Gmail, Calendar, Drive, and location data' is list-based. 'And in the US' is a weak sentence opener that should integrate into prior context. 'without asking users first' is vague compared to 'without explicit consent'.

Google's consumer product connects email, calendar, and location data. Retention periods are longer than OpenAI or Anthropic, and the human review window extends to 3 years. Gmail integration was enabled by default in the US without explicit user consent.

Line 94 — metronomic-rhythm

For personal productivity use, most of this will not matter day to day. For professionals handling sensitive conversations, client data, or regulated information, the picture is more complicated.

Metronomic parallel structure: 'For X, Y. For Z, W.' creates matching cadence. 'most of this will not matter day to day' is a hedged and vague claim. 'the picture is more complicated' is filler that summarizes without adding information. The second sentence should specify the actual complications (Gmail access, retention windows).

For personal productivity, this is mostly irrelevant. For professionals handling client data or regulated information, the Gmail integration and 3-year review window create real risk.

Line 96 — clickbait-heading

Use Gemini's API for Meeting Notes Through Char

Heading reads as an imperative sales instruction ('Use X Through Y') rather than descriptive documentation. This is clickbait/marketing framing. A technical heading should indicate what the section contains, not tell the reader what to do.

## Alternative: Gemini API via Char

Line 98 — antithesis-binary

If you want to use Gemini without routing data through the consumer product, connecting your own Google AI API key through Char gives you API-level data handling: 55-day retention, no model training, no Gmail integration.

Antithesis setup: 'If you want X without Y, then Z.' The opening is a comparison before the solution. 'gives you API-level data handling' is marketing framing (benefits-focused language). The colon-separated list ('55-day retention, no model training, no Gmail integration') is staccato enumeration.

To use Gemini with API-level retention (55 days, no training), connect your Google AI API key through Char instead of the consumer product.

Line 100 — anthropomorphization

Char is an open-source AI notepad for meetings that lets you bring your own API key for Gemini, OpenAI, Anthropic, Mistral, or others. Your meeting notes are stored on your device. You choose which AI provider processes your data, and you can change that decision without rebuilding your workflow.

Anthropomorphization and marketing framing: 'lets you bring your own API key' uses permissive language ('lets you'). 'You choose which AI provider processes your data' is testimonial framing emphasizing user agency. The parallel structure of three sentences creates metronomic rhythm. 'or others' is filler. The final claim about rebuilding workflow is marketing benefit language, not technical description.

Char is an open-source meeting notepad where you bring your own API key. Notes stay on your device, and you can switch providers without rebuilding your workflow.

Line 106 — marketing-framing

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

Marketing / sales call-to-action framing: 'use the AI provider your security team actually trusts' is emotional appeal and testimonial language ('actually trusts' implies trust in the product). The 'and' construction ties the download to a benefit claim. Remove the sales framing and end with the link.

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

LOW — Subtle but Suspicious

Line 15 — conversational-announcement

Here is the full picture.

Conversational announcement that serves no function. The reader can see the table and text that follows. This throat-clearing phrase previews content instead of letting it speak for itself.

Delete this line entirely.

Line 32 — staccato-fragments

There is a floor on this. Even if you turn off Gemini Apps Activity entirely, Google still retains your conversations for up to 72 hours to operate the service. There is no configuration that gives you zero retention on the Google consumer product.

Staccato three-sentence opening ('There is a floor on this. Even if... There is no configuration...') creates artificial emphasis. 'There is' is filler. The three-part progression is metronomic and could be collapsed into two direct sentences.

Even with Gemini Apps Activity turned off, Google retains conversations for 72 hours for operational purposes. You cannot achieve zero retention on the consumer product.

Line 38 — clickbait-heading

The Gmail Access Story

Clickbait heading formula: 'The X Story' is a marketing device that frames information narratively rather than descriptively. A technical heading should identify the section content directly, not invite the reader into a narrative.

## Gmail Integration and Consent

Line 40 — metronomic-rhythm

In late 2025, Google enabled Gemini access to Gmail, Google Chat, and Google Meet by default for US users. By early 2026, Google rebranded this cross-app data flow under the name "Personal Intelligence," which now manages how Gemini connects to Gmail, Drive, Maps, and other Google services. When enabled, Gemini can read your entire email history to power features like email summarization, draft suggestions, and surfacing relevant information across your inbox.

Metronomic list structure: 'Gmail, Google Chat, and Google Meet' followed later by 'Gmail, Drive, Maps, and other Google services' then 'email summarization, draft suggestions, and surfacing relevant information' creates rhythmic enumeration. The final clause is filler that could be cut. The first two sentences are overloaded with dates and rebranding minutiae that distract from the core fact.

Google enabled Gemini to access Gmail by default for US users, then rebranded it as 'Personal Intelligence' in early 2026. This feature lets Gemini read your full email history for summarization and draft suggestions.

Line 48 — clickbait-heading

How to Control Your Data?

Rhetorical question as heading is a clickbait formula. A technical heading should state the topic, not pose it as a question to the reader. This reads like marketing copy inviting engagement rather than documentation.

## Controlling Your Data

Line 58 — conversational-announcement

Human Review in Plain Terms

Phrase 'in Plain Terms' is unnecessary qualification and reads like an author announcement ('I will explain this clearly'). A heading should identify the content; the clarity is shown by the section itself, not promised.

## How Human Review Works

Line 60 — filler-phrase

Google's Gemini Apps Privacy Hub states that trained human reviewers check conversations to evaluate whether responses are low quality, inaccurate, or harmful. This is standard practice across AI providers, but Google's retention policy for reviewed conversations is notably long at 3 years.

Jargon stack: 'trained human reviewers' + 'low quality, inaccurate, or harmful' + 'notably long' uses intensifiers and formal qualifications that pad the sentence. 'Trained' is unnecessary (reviewers are implied to be trained); 'notably long' is a subjective claim that should be backed by comparison, not asserted with intensifier language.

Google's Gemini Apps Privacy Hub states that reviewers check conversations for low quality, inaccuracy, or harm. While standard across AI providers, Google's 3-year retention for reviewed conversations is notably long.

Line 62 — filler-phrase

This means that in practice, a conversation you have today could be read by a human reviewer and retained until 2029, even if you delete it tomorrow.

Filler phrase: 'in practice' and 'could be' are hedging qualifiers that weaken the statement. 'could be read by a human reviewer' is passive and inflated; 'could be reviewed' is direct. The concrete example (2029) is good; the framing is tentative.

A conversation you have today could be reviewed and retained until 2029 even after deletion.

Line 68 — clickbait-heading

How the API Is Different?

Rhetorical question heading is clickbait formula. A technical heading should state what the section contains, not pose it as a question.

## The Gemini API: Different Retention Rules

Line 70 — antithesis-binary

If you use the Gemini API directly rather than the consumer product, the data policy is different. Google retains API inputs and outputs for up to 55 days for abuse monitoring. This data is not used for model training.

Antithesis setup: 'If you use X rather than Y, the policy is different.' The opening comparison is announced before the actual policy is stated. The three-sentence structure creates metronomic rhythm that could be collapsed. 'This data is not used' is a separate affirmation after negation comparison.

The Gemini API retains inputs and outputs for 55 days for abuse monitoring only—not for model training.

Line 72 — metronomic-rhythm

Two newer API features add nuance here. The first is Context Caching: users can store large amounts of data on Google's servers to reduce token costs when making repeated calls against the same content. This cached data has its own TTL (Time-to-Live) setting that you control, and it sits outside the standard 55-day abuse monitoring window. It is a flexible feature, but it means you may have data on Google's servers for longer than the default if you use it.

Metronomic list structure: 'The first is X...' implies more items with parallel structure, creating AI-like enumeration. Filler: 'add nuance here', 'It is a flexible feature, but' hedges the statement. The final sentence repeats the retention concern already established ('longer than the default'). Multiple nested clauses create rhythmic density.

Context Caching lets you store data on Google's servers with your own TTL setting, separate from the 55-day abuse window. This trades reduced token costs for longer data retention than the default.

Line 76 — filler-phrase

For Vertex AI on Google Cloud, Google offers a Zero Data Retention option on certain endpoints. Under ZDR, prompts and responses are not logged or stored beyond what is needed to return the result.

Filler and passive voice: 'Google offers a Zero Data Retention option on certain endpoints' is indirect. 'are not logged or stored' is passive. The acronym explanation 'ZDR' takes extra space. The second sentence could be collapsed into the first.

Vertex AI on Google Cloud offers Zero Data Retention: prompts and responses are not logged beyond what's needed to return the result.

Line 80 — clickbait-heading

How Retention Works for Workspace and Enterprise Plans

Generic phrase 'How X Works' is a common listicle heading formula that doesn't tell the reader what the section argues or contains. A more specific heading conveys the actual difference.

## Workspace and Enterprise: Stronger Data Protections

Line 88 — conversational-announcement

The Broader Context

Vague heading that doesn't signal content ('Broader Context' is filler). The next sentence makes clear the section compares OpenAI, Anthropic, and Google, so the heading should say that directly.

## Comparing the Three Providers

Line 102 — metronomic-rhythm

For teams that need to go further, Char supports fully local models via Ollama. Your conversations never leave your device at all.

Metronomic rhythm: two sentences with parallel structure. 'For teams that need to go further' is a positioning phrase (marketing framing). 'never leave your device at all' uses intensifier repetition ('never', 'at all') for emphasis. 'at all' is filler. Collapse into one sentence.

Char supports local models via Ollama for fully on-device operation.

Powered by Claude Haiku 4.5 with stop-slop rules

Blog Post Review: Humanizer + Stop-Slop

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

Humanizer Check (24 AI writing patterns)

Score: 37/50 (PASS)

High Severity

Line 104 — Pattern: Negative Parallelism (#9)

That is the practical difference between using a consumer AI product and choosing your own stack. The first gives you a privacy policy. The second gives you control.

Classic "this/that" setup with parallel clauses. Three-part binary contrast structure is a strong AI tell.

Suggested rewrite:

Consumer AI products give you a privacy policy. Your own API key gives you control.

Line 13 — Pattern: Negative Parallelism (#9)

When you evaluate OpenAI or Anthropic, you're asking what they do with your AI conversations. When you evaluate Google, you're asking what they do with your AI conversations and everything else they already know about you.

Anaphoric "When you evaluate X / When you evaluate Y" with repeated clause creates AI rhetorical cadence.

Suggested rewrite:

Google's risk profile differs from OpenAI or Anthropic because Gemini combines AI conversation data with email, location, search, and calendar data Google already has.

Medium Severity

Line 15 — Pattern: Inflated Symbolism (#1)

Here is the full picture.

Vague meta-commentary. "Full picture" is abstract and adds no information.

Suggested fix: Delete this line entirely.

Line 34 — Pattern: Vague Attribution (#5)

Here is the part that catches most people off guard:

Unsubstantiated claim about what surprises users + throat-clearing opener.

Suggested fix: Delete "Here is the part that catches most people off guard:" and lead directly with the factual claim.

Line 38 — Pattern: Inflated Symbolism (#1)

The Gmail Access Story

"Story" is unnecessarily narrative framing for a heading.

Suggested rewrite: Gmail Access and Default Settings

Line 46 — Pattern: Inflated Symbolism (#1) + Negative Parallelism (#9)

Google's position is that this data is used only to power features for individual users, not to train its public AI models. That is a meaningful distinction. But for anyone evaluating Gemini as a provider for sensitive work, the default-on access to your full email history is worth knowing about upfront.

"That is a meaningful distinction. But..." creates setup-and-counter rhetorical pattern.

Suggested rewrite:

Google states this data powers individual features, not model training. For anyone handling sensitive work, the default-on Gmail access matters.

Line 58 — Pattern: Meta-commentary

Human Review in Plain Terms

"In plain terms" signals the writer's approach rather than describing content.

Suggested rewrite: Human Review Details

Line 64 — Pattern: Vague Attribution (#5)

One area that often gets overlooked:

Claims knowledge of what "often gets overlooked" without evidence.

Suggested rewrite: Start directly with the Gems detail.

Line 66 — Pattern: Inflated Symbolism (#1)

That is practical advice, but it places the compliance burden squarely on users rather than on the product design.

"Squarely" is an intensifier; formal analytical tone reads as AI.

Suggested rewrite:

This shifts compliance responsibility from Google's product design to individual users.

Line 72 — Pattern: Inflated Symbolism (#1)

Two newer API features add nuance here.

"Add nuance" is meta-commentary about complexity.

Suggested rewrite:

Two newer API features affect retention differently.

Line 88 — Pattern: Inflated Symbolism (#1)

The Broader Context

Generic transition heading.

Suggested rewrite: Comparing Gemini to Other AI Providers

Line 90 — Pattern: Inflated Symbolism (#1)

OpenAI and Anthropic each had a single data retention policy story to tell. Google's story is more layered...

"Story to tell" and "layered" are unnecessarily narrative.

Suggested rewrite:

OpenAI and Anthropic have straightforward retention policies. Google's is more complex because Gemini integrates with Gmail, Calendar, Drive, and location data.

Low Severity

Lines 28, 48, 68 — Question marks on section headings (What Gemini Stores by Default?, How to Control Your Data?, How the API Is Different?) — minor style inconsistency. Consider removing the question marks.

Line 32 — There is a floor on this. — Slightly abstract metaphor. Consider: "Google sets a minimum retention period."

Stop-Slop Check (phrases, structures, rhythm)

Score: 37/50 (PASS)

Banned Phrases

Line 15 — Throat-clearing opener

Here is the full picture.

"Here is..." construction is throat-clearing before the content. Delete entirely.

Line 34 — Throat-clearing + meta-commentary

Here is the part that catches most people off guard:

Announces the interesting part instead of stating it. Cut and lead with the fact.

Line 74 — Filler phrase

...but it is worth knowing if you are using the voice API for sensitive conversations.

"It is worth knowing" is meta-commentary about importance. Rewrite to direct statement.

Line 46 — Emphasis crutch

That is a meaningful distinction.

Tells readers something is important instead of showing why. The distinction is already clear from context. Delete.

Line 60 — Adverb

...Google's retention policy for reviewed conversations is notably long at 3 years.

"Notably" is unnecessary emphasis. The 3-year figure speaks for itself.

Line 66 — Adverb

...it places the compliance burden squarely on users...

Empty emphasis. Cut "squarely."

Structural Cliches

Line 104 — Binary contrast

The first gives you a privacy policy. The second gives you control.

Classic "The first X. The second Y." formulaic construction. Collapse into one direct comparison:

A consumer product gives you a privacy policy; your own stack gives you control.

Line 13 — Binary contrast

When you evaluate OpenAI or Anthropic... When you evaluate Google...

Parallel "When you X / When you Y" creates false drama. State the point directly.

Line 36 — Binary contrast

Google says it tries to... But if you typed personal information...

"Tries to X. But if Y, then Z." is a textbook AI rhetorical setup. Merge into one statement:

Google attempts to anonymize reviewed conversations, but personal information you typed remains visible to the reviewer.

Line 66 — Binary contrast

That is practical advice, but it places the compliance burden...

Setup-and-counter move. State the consequence directly:

This shifts compliance responsibility from product design to user behavior.

Rhythm Patterns

Line 82 — Metronomic / staccato fragments

...the rules are more protective. Prompt content is not used... Human review... does not happen... Workspace admins can shorten...

Three parallel sentences with identical negation structure ("is not used", "does not happen"). Combine:

Workspace Enterprise customers get stronger protections: no model training without permission, no human review without consent, and admin control over storage duration.

Line 84 — Metronomic rhythm

On HIPAA: ... On GDPR: ...

Parallel "On X:" openers create formulaic structure. Vary the sentence starters.

Passive Voice

Line 70 — Google retains API inputs and outputs for up to 55 days for abuse monitoring. This data is not used for model training.

"This data is not used" hides the actor. Fix: "Google does not use this data for model training."

Summary

Both checks pass with a score of 37/50 each, just above the 35/50 revision threshold. The post's core strength is its specificity — real case names, concrete retention periods, exact URLs, and technical nuance. The weaknesses are structural:

1. Binary contrasts (lines 13, 36, 46, 66, 104) — the "X. But Y." pattern appears throughout
2. Throat-clearing openers (lines 15, 34) — "Here is the full picture" / "Here is the part that catches..."
3. Meta-commentary (lines 46, 58, 64, 72, 74) — sentences that announce importance rather than show it
4. Closing paragraphs (lines 96-106) — shift to marketing voice with binary contrast ending

The technical middle sections (API details, Workspace policies, human review) are strong. The intro, transitions, and closing are where AI patterns concentrate. A targeted edit pass on those areas would improve both scores significantly.