1. What we hold ourselves to
People decide
AI drafts, summarises, suggests and scores. Where a result affects someone else, a person decides what happens — unless you have switched on an automation yourself, with limits you set.
We say when it is AI
You should never have to guess whether you are talking to, hearing, or being assessed by AI.
We do not invent
If a feature cannot measure something, it leaves it blank. It does not guess and present the guess as fact.
Your content is not training data
We do not train AI models on your content, and the AI providers we use are not permitted to either.
Coaching signals are yours
Feedback drawn from your voice or your camera is shown to you, for your own practice. It is never used to rate you for an employer.
A person is always reachable
You can ask one of us to review any AI result about you, and to explain how it was reached.
2. What each product does with AI
Every AI feature in the Suite, in plain terms: what it does, which provider does the work, and who stays in charge of the result.
The Mirror™
themirrorai.appCommunication coaching
Reads the emails, messages and meeting transcripts you choose to share and shows you how you come across as a leader.
ProviderAnthropic
Who decidesCoaching for you. Nothing is sent or changed on your behalf.
Voice reflection
Turns a recording of your voice into text and reflects patterns such as pace and tone back to you.
ProviderAssemblyAI (speech to text), Anthropic
Who decidesShown only to you.
On-camera presence
Looks at a few still frames from your webcam and comments on framing and how open and engaged you appear. The frames are not stored, and it does not try to detect emotions.
ProviderAnthropic
Who decidesShown only to you.
Slack and Teams
If you connect Slack or Teams, looks at a message only when you send it to Mirror (with /mirror in Slack, or in your one-to-one chat with Mirror in Teams), and can send you one short coaching note a week that you can reply to. It does not read your channels or other conversations.
ProviderAnthropic
Who decidesOnly you see the answers. Mirror keeps the scores, not your words. Replies to the weekly note are kept for 30 days. The weekly note is off until you turn it on.
Team view
Shows a team owner averages across a group, and only when the group is large enough that no one person can be picked out.
ProviderAnthropic
Who decidesThe owner sees group averages only — never one person's results.
Read aloud
When you press Listen, reads one of your Letters from your Mirror or voice reflections back to you in a synthesised voice. The audio is kept in your own private storage so you can play it again.
ProviderElevenLabs (the voice)
Who decidesOnly for you, and only when you ask.
When Mirror analyses a recorded meeting, AssemblyAI is told to delete the audio and transcript as soon as the text has been produced.
The Rehearsal Studio™
therehearsalstudio.appPractice conversations
An AI plays the other side — an interviewer, a guest, a prospect or a colleague — out loud, in real time.
ProviderAssemblyAI (live speech to text), Anthropic (the conversation), ElevenLabs (the voice)
Who decidesA practice session you start and end.
Debrief and scores
Scores your session against the module's criteria from what you said, and picks out moments worth rewatching.
ProviderAssemblyAI, Anthropic
Who decidesFeedback for you, emailed only to you.
On-camera presence
Looks at a handful of still frames for eye contact, framing and energy. It never comments on appearance or identity, and the frames are not stored.
ProviderAnthropic
Who decidesShown only to you.
Your session recording is deleted automatically 15 days after the session unless you choose to keep it, and you can delete it yourself at any time.
The Greenroom™
thepodcastgreenroom.appAsk the Archive
Answers members' questions from the show's episode catalogue.
ProviderAnthropic
Who decidesAn answer on screen, labelled as written by AI.
Interview practice
An AI stand-in plays a booked guest, using the guest's published bio, so members can practise their questions — then gives feedback on those questions.
ProviderAnthropic
Who decidesPrivate practice, labelled as an AI stand-in.
Member introductions
Suggests members who might get on with each other.
ProviderAnthropic
Who decidesAn admin approves every introduction before any member sees it.
Show notes and member emails
Drafts show notes and emails to members for the show's admins.
ProviderAnthropic
Who decidesDrafts. An admin edits and sends.
The Broadcast™
thebroadcastai.appYour website
Writes the pages of your show or business site from your answers, and keeps show notes and your homepage current as new episodes arrive.
ProviderAnthropic
Who decidesPublished to your site. You can edit or rewrite any page, and pages you have edited by hand are not overwritten.
Episode and guest tools
Show notes, repurposed posts, a sponsorship pitch, and promotion for your guests.
ProviderAnthropic
Who decidesYou approve anything said about a guest before it is public. Everything else can be edited.
Tools for your visitors
Answers visitors' questions about your show or business, tailors your headline to where visitors came from, and responds to quiz and valuation forms.
ProviderAnthropic
Who decidesVisitors are told the answers are written by AI.
The Production™
theproduction.appGuest pre-interview
An AI producer holds a short spoken pre-interview with a guest in their browser. Guests are told it is AI, and that it is recorded, before it starts.
ProviderAnthropic (the conversation), ElevenLabs (the producer's voice), AssemblyAI (live speech to text)
Who decidesA conversation only. It does not decide anything.
Screening summary
Summarises the pre-interview and scores fit against the show's own criteria, with every point linked to the exact moment it came from.
ProviderAnthropic, AssemblyAI (optional audio analysis)
Who decidesThe host decides. A score never books or turns down a guest on its own.
Prep and production
Research briefs, interview guides, show notes, articles, sponsor pitches, audio clean-up and summaries of recorded episodes.
ProviderAnthropic, Auphonic (audio clean-up)
Who decidesDrafts for the host. Hosts see any promotional email before it goes to a guest.
Recording companion
A private second screen for the host while they record: it listens through the host's microphone, suggests follow-up questions, and shows what is left of the planned questions and when the sponsor read is due. The host confirms they have told the guest before it listens.
ProviderAssemblyAI (live speech to text), Anthropic
Who decidesSuggestions for the host only. Off unless the account switches it on. The words heard are not kept unless the host chooses to keep the transcript.
Pre-interview recordings are deleted after the show's retention period — 30 days by default, after a 7-day warning. Hosts can change the period.
The Forge™
theforgeai.appMeeting follow-up
Reads the transcripts of your recorded Zoom meetings and turns them into tasks, decisions, notes and suggested contacts.
ProviderAnthropic
Who decidesTasks and notes are created for your team. Nobody is added to your contacts unless you choose Add.
Follow-up drafts
After a recorded Zoom meeting, drafts a follow-up email or a short proposal outline from that meeting's transcript, tasks and decisions.
ProviderAnthropic
Who decidesA draft saved on the meeting for you to edit. Forge never sends it.
Ask Forge
On plans that include it, answers questions about your company's own Forge records, and drafts from them. It is given your projects, tasks, decisions, meeting topics and action items, contacts and deals, wiki pages and standups, but never a meeting transcript.
ProviderAnthropic
Who decidesAn answer on screen for you. We do not store the conversation.
The Signal™
thesignalai.appWeekly briefing
Reads the week's reviews, news and mentions about you and writes a briefing, with an optional read-aloud version.
ProviderAnthropic, ElevenLabs (read-aloud voice)
Who decidesEmailed to you.
AI visibility
Asks AI assistants the questions a potential client would ask about your field — cold, with no hint who is asking — and records whether you were named and who was recommended instead. Four of the five questions do not name you; the fifth asks what the assistant knows about your business. Claude and ChatGPT answer each question twice, once from their own knowledge and once after searching the live web. Perplexity always searches.
ProviderAnthropic, OpenAI (ChatGPT), Perplexity
Who decidesA weekly report for you. Nothing is published unless you approve a correction and choose to send it to your own Broadcast site.
Ask The Signal
Answers your questions from your own history of mentions and briefings.
ProviderAnthropic
Who decidesAn answer on screen for you.
Reply drafts
Drafts a reply to a review, post or journalist, following your compliance rules.
ProviderAnthropic
Who decidesCopy only. Nothing is posted for you.
Mention ratings
Marks each mention positive, neutral or negative — from star ratings, or from DataForSEO's own reading of the page for web and news mentions — and uses that in your Signal Score.
ProviderDataForSEO (web and news mentions)
Who decidesAutomatic. A Thursday mid-week email, which you can turn off in Settings, lists up to five of the week's negative or most relevant mentions. Your weekly briefing takes the ratings into account.
The Outpost™
theoutpostai.appProspect research
Finds prospects on the public web that match the buyer you describe, and scores how well they fit.
ProviderAnthropic, Firecrawl (web search)
Who decidesA list for you. Prospects below your fit threshold are left out.
Outreach
Drafts first messages in your voice.
ProviderAnthropic
Who decidesDrafts always wait for you. Nothing reaches a prospect unless a person sends it. After each update a shortlist picks out the strongest drafts, but it sends nothing itself.
Replies
Reads replies from prospects, marks them positive, neutral or negative, and suggests a response.
ProviderAnthropic
Who decidesYou send any response.
Market and community watch
Watches competitors, market signals and community posts, and drafts replies and content.
ProviderAnthropic, Firecrawl (web search)
Who decidesDrafts for you. Nothing is posted to LinkedIn until a person presses Publish on that post.
Help assistant
Answers questions about setting up and using the app, using only the app's own help text. It can't see your account.
ProviderAnthropic
Who decidesIt can't change or send anything.
theunfairai.com
theunfairai.comConcierge
Answers questions about what we build and what it costs, using only what this site publishes.
ProviderAnthropic
Who decidesIf you leave your details, one of us reads the conversation and replies personally.
AI Readiness Check
Writes a short read on where AI would and would not help your business, from your five answers. Your name and email are not sent to the AI.
ProviderAnthropic
Who decidesWe see your answers and the read, and follow up personally.
3. Who processes your data
Every company named as a provider above is listed here. Anthropic, AssemblyAI, ElevenLabs and Auphonic each have a signed data processing agreement with us, and none of them may use your content to train their models. OpenAI, Perplexity, DataForSEO and Firecrawl are each used by a single product and are sent far less, so for each one the table says exactly what it is sent and why.
| Provider | What it does for us | Trains on your content? | How long it keeps it |
|---|---|---|---|
| Anthropic (Claude) | Writing, analysis, coaching, conversation and scoring, in every product. | No. Anthropic does not train on content sent through its commercial API. | A limited period under its commercial terms, then deleted. |
| AssemblyAI | Speech to text and audio analysis in The Mirror, The Rehearsal Studio and The Production. | No. We have opted out of its model-improvement programme. | Up to 30 days. The Mirror deletes each job as soon as the text is produced. |
| ElevenLabs | Spoken voices in The Rehearsal Studio and The Production, read-aloud briefings in The Signal, and read-aloud letters and reflections in The Mirror. | No. Data use for training is switched off. | A limited period under its terms, then deleted. |
| Auphonic | Cleans up the audio of recordings in The Production. | No. Covered by our data processing agreement. | Only as long as it takes to process the file. |
| OpenAI (ChatGPT) | The Signal's weekly AI visibility check. It is sent the five questions written for your account (four about your field that do not name you, and one asking what it knows about your business) and nothing else about you. It answers each one twice: once from its own knowledge, once after a web search. | No. OpenAI's API terms say it does not train on content sent through its API. | We switch off its option to store each answer. Its terms let it keep requests for up to 30 days to check for abuse. |
| Perplexity | The same weekly check in The Signal, always with a web search. It is sent the same five questions and nothing else about you. Perplexity may have another company's AI model write the answer. | No. Perplexity's API terms say it does not train on content sent through its API. | As its own terms allow. |
| DataForSEO | Web and news mentions in The Signal. Once a week it is sent each keyword you ask The Signal to watch, which usually include your name or business name, and it returns the pages that mention them, with its reading of each page's tone. | No promise either way. It is sent only your keywords. | As its own terms allow. |
| Firecrawl | Web research in The Outpost. It is sent web searches written from your description of what you sell and who buys it, the names of companies being researched, and the web addresses of competitors you track, and it returns the pages it finds. | No promise either way. It is sent only those searches and web addresses. | As its own terms allow. |
4. What we will not build
- AI that decides who gets a job, a loan, housing, insurance, healthcare or a place in education.
- AI that reads people's emotions for an employer, a school, or anyone other than the person themselves.
- Results drawn from someone's voice or face that are shared with anyone other than that person.
- AI that pretends to be human, or passes itself off as a real person.
- A copy of a real person's voice made without their written consent.
- Recording or monitoring anyone without telling them.
- AI that follows instructions hidden inside the emails, reviews or web pages it reads.
5. Your right to a person
If one of our products has produced a result about you — whether you are our customer or someone our customer works with — you can ask us to:
- have one of us review an AI result about you, and correct it if it is wrong
- explain, in plain terms, how that result was reached
- correct or delete the information a product holds about you
If the information came from one of our customers — a show host who booked you, or a business that contacted you — we will pass your request to them and help them answer it, because it is their record.
How to ask
Email support@theunfairai.com with the product name and what you would like us to look at. We reply within five business days and resolve it within thirty.