Building it takes less effort than proving it. Personify trains your Second Self only on the content you upload, and it answers your clients around the clock in your voice, using your frameworks. The harder part is evidence that it holds up under real questions. These tests are your quality control pass.
If you are still deciding whether a Second Self can sound like you at all, read the complete voice cloning guide for coaches first. This post assumes your Second Self already exists.
7
tests to run before a paying client sees it
5
answers sent blind to people who know you
3 or 4
reviewers who know how you write or speak
1 to 5
the rating scale in the client survey
Key takeaways
- The blind message test is the single cleanest check: if people who know you cannot spot your Second Self's answers, the voice is working.
- The seven tests are the blind message test, refusals, tone under pressure, framework use, repeat questions, phrasing, and a client survey.
- A good Second Self knows when to stop. The refusal test matters as much as the accuracy test.
- Every miss goes on one list, and you fix that list before launch.
Why voice is a launch blocker
Clients rarely say the cadence was slightly off. They say "that didn't feel like you" and quietly stop trusting the tool. If you charge for access, a gap in voice costs you more than a gap in features.
So treat launch like a product release. You would watch a course module back before you sold it, and your Second Self deserves the same pass. The seven tests move from objective checks (can people who know you tell the difference) to subjective ones (do your clients feel it is you).
Objective: can people who know you tell the difference?
- Test 1The blind message testDid you write this, or did a tool?
- Test 2The refusal testWhere your expertise ends
- Test 3The upset client testTone under pressure
- Test 4The framework testYour method, not generic advice
- Test 5The repeat question testConsistency across wordings
- Test 6The phrasing testYour verbal fingerprint
- Test 7The client surveyDo your clients feel it is you?
Subjective: do your clients feel it is you?
| Test | What you run it on | How many |
|---|---|---|
| 1. Blind message test | Answers to common client questions, sent to people who know how you write or speak | 5 answers, 3 or 4 people |
| 2. Refusal test | Prompts outside your scope, written down before you start | 5 prompts |
| 3. Upset client test | Difficult scenarios, plus one rude message | 3 or 4 scenarios, 1 rude message |
| 4. Framework test | Common client scenarios, scored against your signature frameworks | 5 to 10 scenarios, 3 to 5 frameworks |
| 5. Repeat question test | The same question in different wordings | 5 wordings |
| 6. Phrasing test | Answers read against your list of tells | 10 answers |
| 7. Client survey | Questions rated from 1 to 5, plus one open comment box | 3 or 4 questions |
Every number comes from the test descriptions below.
Test 1: The blind message test
Take five of your Second Self's answers to common client questions and remove anything that reveals the source. Send them to three or four people who know how you write or speak. Ask one thing: did you write this, or did a tool?
This is the cleanest check you can run. If your reviewers guess "you" most of the time, you are passing on the dimension that matters most. If they hesitate or guess wrong, note which answers tripped them up. Usually it is one or two phrasings, and the rest holds.
Test 2: The refusal test
A Second Self that answers everything is a liability. A good one knows where your expertise ends, so test that on purpose.
Write down five prompts outside your scope before you start. Push toward medical and legal questions, and ask what a client in crisis might ask. Your Second Self should not improvise a therapist's answer or hand out legal advice. It should hold the line and point the person to a qualified human.
Good behaviour sounds like this: "That is outside what I can help with here, and it is worth speaking to a professional." If it tries to answer anyway, add that prompt to your fix list. Your Second Self supports the questions you already answer. It is no substitute for crisis care.
Test 3: The upset client test
Anyone can sound like you on an easy question. The real test is a charged one. Run three or four difficult scenarios through it, such as a student who feels stuck or a client who thinks your method has stopped working. Add one rude message for good measure.
Look for two things. First, check that the tone holds without turning defensive or generic. Second, check that it still sounds like you under that pressure. Many experts have a particular way of staying calm and direct with a frustrated client, and your Second Self should carry it.
Read the answers out loud. If a sentence is one you would never say to an upset client, mark it. Tone drift under pressure is fixable once you can point at the exact line.
Test 4: The framework test
Your clients pay for your method. Your Second Self has to apply your frameworks and avoid the average of everything on the internet.
Pick five to ten common client scenarios, run each one through it, and compare the answer with how you would handle it. Check whether it reaches for your steps and terminology, or falls back on advice any chatbot could give.
A quick way to score it:
- 01
List your three to five signature frameworks.
- 02
Ask a question that should trigger each one.
- 03
Mark each answer as applied my framework, applied generic advice, or mixed.
If most answers draw on your own material, the grounding is working. Generic advice creeping in points to a gap in your source material, and you close it by adding the missing content. For more on how it learns your approach, see how to create an AI version of yourself for coaching.
Test 5: The repeat question test
Ask the same question five times in five different wordings. Change the order of the words, make one version casual, and make another long and rambling.
You are checking consistency. Your advice on a given problem should come out recognisably the same whichever way the client phrases it. If one wording gets your method and another gets a generic answer, the grounding is thin on that topic. Note the topic and add content that covers it.
This test also catches the opposite problem. If every answer opens with an identical sentence, the reply has become a template, and clients will spot the pattern quickly.
Test 6: The phrasing test
Every expert has a verbal fingerprint: signature phrases, the way you open a reply, the words you reach for, and the ones you avoid. This test checks whether your Second Self uses yours or slides back into default AI phrasing.
Make a short list of your tells before you start:
- Phrases you use constantly, such as "let's get specific" or "start with the next small step", or whatever yours are.
- Words you would never use with a client.
- Your usual way to open and close a conversation.
Then read ten answers against the list. Look for your phrases showing up naturally, and for filler you would never say. If it keeps reaching for words that are not yours, write them down. That gives you something exact to correct, where "it doesn't sound like me" gives you nothing to act on.
Test 7: The client survey
The first six tests have you grading your Second Self. This one brings in the only group whose opinion decides the outcome, your clients. Before a full launch, share it with a small group of real students and collect structured feedback.
Keep it short: three or four questions rated from 1 to 5, plus one open comment box.
| Question | Scale |
|---|---|
| Did the answers sound like [your name]? | 1 to 5 |
| Was the answer useful? | 1 to 5 |
| Did it fit how [your name] works with clients? | 1 to 5 |
| Anything that felt off? | Open text |
The numbers show where you stand and the comments show why. If the voice scores high but usefulness lags, the voice is right and the content needs work. If it is the other way round, the tone needs tuning. Either way you launch with evidence.
If your Second Self speaks aloud
Text is the default on every Personify plan. Spoken voice is different. Personify's pricing page says voice cloning is done for you only, because a convincing voice needs a supervised recording session and hands on tuning. If your build includes it, add one more step to the blind test: send a short audio clip and ask the same question. Listeners who flag something robotic give you a specific clip to send back for refinement.
Working through the gaps
Treat every gap as a punch list item. If you built your Second Self yourself, correct the answers that missed and add source content wherever a test exposed a hole. If Personify built it for you, send the list back and ask for a refinement. For a fuller picture of how it is assembled, see how an AI coaching clone works.
Step 1
Run the seven tests
Step 2
Every miss goes on one list
Step 3 · fix it
Built it yourself: correct the answers that missed and add source content.
Personify built it: send the list back and ask for a refinement.
Step 4
Launch with evidence
These tests check fidelity and promise nothing more. Your Second Self handles the questions you answer over and over, which leaves your time for the work that needs you.
The blind test and the refusal test together show what it can carry and where it should step back.
Frequently asked questions
How do I know if my AI clone sounds like me?+
Run the blind message test. Send five answers with the source labels removed to people who know how you write and speak, and ask whether you wrote them or a tool did. If they mostly guess you, it is passing. Wrong guesses point to the exact answers that need refinement.
What if my reviewers split 50/50 on the blind test?+
A split means the voice is close but not settled. Look at which answers drew the "tool" votes and ask those reviewers what gave it away. Fix those phrasings, then run the test again with fresh answers and, if you can, new reviewers who have not seen the first set.
How often should I re-run the tests after adding content?+
Run the blind message test and the phrasing test again whenever you add a batch of new source material, because new content can shift how your Second Self answers. Re-run the refusal test any time you widen what it covers. A full pass of all seven is worth repeating before any major change to how you sell or deliver your work.
Verify it, then launch it
Run the seven tests, collect the evidence, and fix what misses. You then launch knowing how your Second Self performs on your own questions.
Start free and build yours, then run these tests yourself. Personify Pro is $39/month billed monthly, with unlimited messages. If you would rather have it built for you, done for you work is priced on a custom quote after a free discovery call.
Free tier available. No credit card needed.