Personify Deflection and Voice Figures, Defined

The figures, defined

Question deflection

• What it counts: the share of student direct messages a Second Self answers without the creator stepping in.
• Scope: one career coach's Second Self. Our homepage reports that it "Deflected 73% of student DMs".
• Current figure: 73%, for that single deployment.

The unit matters here. A deflected message is a direct message that reached a reply without the creator typing one. Conversations and satisfied students are different units, and neither is what the figure counts. A student can send three messages in one conversation, and each of those messages is its own count. If you read "73%" as "73% of students were helped", you have changed the unit, and the number no longer means what the homepage says.

The homepage gives the 73% as a single figure with no start date, end date, or message count attached. That is why we call it one coach's result and never describe it as typical. Use it to learn what deflection means, and look to your own conversation log to learn what it would mean for you.

Voice likeness

• What it counts: how closely the synthetic voice matches the creator's real voice.
• Scope: the done for you service, which adds voice cloning to a Second Self. The self-serve Free and Pro plans do not include it.
• Current figure: 97% likeness, as stated on our pricing page.

The figure belongs to a specific build. The pricing page lists a "Professional voice clone (97% likeness) trained on hours of your recordings" as part of every done for you build, so the number describes a clone trained on a creator's own recordings. Personify has not published the measurement method behind the 97%, and the pricing page, which states it without a date, is its only source.

What you can check yourself is the listening review before launch. The pricing page describes a 60-minute discovery call to capture your voice, frameworks, and audience, and a review by you before launch, with live deployment around day 14. You hear the result before your audience does, and your ear is a better judge of your own voice than any percentage.

A worked example: reading one client's results

Take the same career coach behind the 73%. Her case study shows a usage table headed "Usage statistics (Oct 1-9, 2025)" with 1,056 total conversations in that window and an average duration of 11 minutes. It also lists weekly direct messages handled, at about 500 handled manually before her Second Self and 2,000 or more handled automatically after it.

Reading those figures carefully turns up four differences.

First, the windows. The conversation count covers nine days in October 2025, and the 73% carries no dates at all. You cannot divide one by the other, so the two numbers do not combine into a rate.

Second, the units. The case study counts conversations in one table and direct messages in another. A conversation can hold many messages, so 1,056 conversations says nothing about how many messages were deflected. The weekly handled count is a third unit again: messages per week, before and after.

Third, the scope. Every number in that case study belongs to one coach with her own audience and pricing. The same page lists her response time before her Second Self as "Hours to days", which tells you the starting point she measured against, and your starting point will differ.

Fourth, the headline. The case study is titled "2,000 DMs to zero manual replies". That headline describes weekly message counts, 2,000 or more handled automatically and none handled manually, and the same page opens its story with 2,000 LinkedIn direct messages arriving in 48 hours after one viral post. The 73% on the homepage is a share of student direct messages. A weekly count and a share are different units, so neither converts into the other without the total message volume behind it. Read each one as the page that states it defines it.

The before column of the same case study gives the starting point for the 73%. It lists weekly direct message volume of about 500, more than 12 hours a week spent on admin, and a student capacity of 500 to 600. Those are the conditions the deflection figure was produced under. A coach with a smaller inbox starts from a different baseline, so the same percentage would free up fewer hours, and a coach with a larger inbox would free up more. A rate says how much of the inbox was handled. The inbox size tells you how much that matters.

The case study itself carries the caveat that matters most. It says: "What a clone earns depends on your audience, pricing and offer." The same applies to deflection. A coach whose students ask the same ten questions every week will see a different rate from one whose students ask something new each time.

How to check any number

Use this sequence on our pages, and on anyone else's.

1. 01

   Find the unit. Is it a message, a conversation, a student, or a sum of money? If the page does not say, the figure is not ready to use.

2. 02

   Find the scope. Is it one client, a group of clients, or every account? A figure from one client is an example. A forecast needs a wider base than that.

3. 03

   Find the period. A figure with no dates cannot be compared with anything. Where a figure carries no month and year, keep it out of any calculation.

4. 04

   Find the source. For a claim about the world outside the vendor, the source should be a named page you can open. For a claim about the vendor's own platform, the source should be the page that states it, read by you on the day you rely on it.

5. 05

   Check the date of the page. Where a page shows when it was last updated, read that date. A figure on a page last updated two years ago deserves a second look.

Run on the 73%, the sequence gives a message as the unit, one career coach as the scope, no stated period, and the homepage as the source. On the 97% it gives the likeness of a synthetic voice to a real one as the unit, the done for you service as the scope, and the pricing page as the source, again with no stated period. Neither is a forecast for your business. Both are definitions you can test against your own audience and your own voice.

What a figure does not tell you

Your results depend on how repetitive your students' questions are and how much of your teaching has been loaded in. The pricing page says done for you includes "Deep multi-format content training (books, courses, PDFs, videos, frameworks)" and ongoing retraining as you add content, so the content you bring shapes what your Second Self can answer. Voice likeness describes a service build and says nothing about how good the answers are.

A high deflection rate counts the messages answered without you. Whether each answer was right is a separate question, and your own reading of what your Second Self tells your students is the way to find out. Every conversation in the product is logged, so you can read the answers yourself, and we think you should, especially in the first weeks. Personify publishes no figure for satisfaction or learning outcomes.

When you test a Second Self on your own material, three readings are worth taking in the first weeks. Read a sample of the logged conversations and mark each answer as matching your own teaching or not. Note which questions repeat, because repeated questions are where deflection comes from. Listen for anything the Second Self says that you would never say, and, on a done for you build, tell your success manager, since the pricing page lists ongoing retraining as part of the build. On self-serve, add the missing material yourself. Those readings give you your own version of the figures on this page, built on your own audience.

How often this updates

We review this page when the figures change. The update date in the byline changes whenever a definition changes, so you can tell when they last moved.

A note on outside research

Where our articles cite outside research, such as coaching efficacy or market size, we link the original study or release directly and leave this page out of it. This page covers only Personify's own platform data.