Voice Profiles

Per-domain writing voice profiles that ensure content sounds authentic to each brand.

What Voice Profiles Do

Every domain in Studio has a voice profile stored in aeo_pipeline_authors. When the Write Article skill generates content, it applies the profile's rules to match the brand's natural writing patterns - sentence structure, vocabulary choices, and tone markers.

Without a voice profile, content reads like generic AI output. With one, it mirrors the patterns found in the brand's actual writing samples.

Source Materials

Voice extraction accepts several types of writing samples:

LinkedIn PDF

Download the profile as PDF from LinkedIn. Rich source of professional voice and expertise signals.

Blog posts

Existing articles from the brand's website. Best for capturing established content patterns.

Writing samples

Any text the author has written - emails, newsletters, case studies. 500+ words recommended.

Use writing samples from varied contexts - a blog post, a LinkedIn update, and a case study give a more complete picture than three blog posts on similar topics.

8 Style Dimensions

Voice extraction analyzes writing samples across 8 dimensions to build a comprehensive style fingerprint.

Sentence lengthAverage word count, mix of short and long sentences, cadence patterns

Vocabulary levelReading grade level, technical term frequency, jargon preferences

Punctuation habitsEm-dash vs comma usage, semicolons, parenthetical asides, exclamation frequency

Paragraph rhythmAverage paragraph length, single-sentence paragraphs, transition patterns

Transition stylesConjunction preferences, linking phrases, section flow techniques

Rhetorical devicesQuestions, analogies, metaphors, anaphora, rule of three

Tone markersFormality level, humor indicators, empathy signals, authority cues

Forbidden patternsOverused words, cliches, patterns that feel inauthentic to the brand

Profile Fields

The extracted profile is stored with the following fields on aeo_pipeline_authors, keyed by domain.

tonestringOverall tone descriptor (e.g., "warm and authoritative", "casual but informed")

do_rulesstring[]Patterns to follow - writing habits that define the voice

dont_rulesstring[]Patterns to avoid - anti-patterns that feel inauthentic

brand_stancestringHow the brand positions itself on key topics

content_guidelinesJSONBExtended guidelines including target audience, style notes, and preferences

target_criteriastring[]Specific criteria or goals for content (e.g., "emphasize original data", "use question headings")

Managing Profiles

Profiles are managed through the aeo_upsert_client_author MCP tool. Each domain has one profile. Running voice extraction again on the same domain updates the existing profile rather than creating a duplicate.

You can also manually edit profile fields in Studio by asking Claude to update specific rules. For example: "Add a dont_rule: never use the phrase 'in today's world'".

Tips for Better Profiles

Provide at least 500 words of writing samples. More text gives the analysis more patterns to work with.
Use samples from different contexts - blog posts, social media, case studies - for a more complete voice fingerprint.
Review the generated do_rules and dont_rules. Remove any that feel off and add rules you know are important for the brand.
Re-run voice extraction after publishing several articles. The larger body of work produces a more accurate profile.
Check the best practices page for AEO-specific patterns that should be included in target_criteria.