Pick your readers
Say who they are: age, education, how well they know the topic, and where and why they read. That sets how much they know, how much they care and how fast they read.
Simulated reader roundtable
Read the room
before you walk in.
Easy to generate. Hard to resonate.
Test your memo, pitch or deck before the meeting that counts. Pick the panel you will face: the board, an investor, your customers, a copy editor. Each simulated reader reads it for their own role and at their own pace, some to the end and some only to the sentence where they give up. Then they sit around one table and tell you where it held them, where it lost them, and what they would push back on.
- Sentence by sentenceAttention, comprehension and memory advance one sentence at a time, reproducibly
- Each with their own lensA role sets what they read for; age, situation and motivation set how fast they read and how far they get
- Grounded in the textEvery remark quotes what that reader actually read
Core
Message Bench
Paste a draft and seat three to five readers. Watch each finish it or put it down at their own pace, hear them speak and answer one another, then take the moderator's edits in order of priority.
Go to the bench ProfilesImport and fine-tune readers
Start from the five built-in readers or import agent-profile-v1 files. Move a reading parameter and the engine re-reads the piece as that reader, live.
Open the profile library SettingsPreferences and saves
Choose the default seats and the replay speed. Every roundtable is saved on this device and replays without calling a model again.
Open settingsMethod
How it works
They read, sentence by sentence
Each reader's attention and understanding rise and fall as they go, and some stop early. The same draft and the same readers always give the same result.
They tell you how it went
Each one says what held them and where they lost the thread, quoting only the part they actually read.
Good to know
- Simulated.The seats hold simulated readers. Names and titles such as Ashley or Dr. Russell are fictional and stand for no real person.
- Traceable.Every quote comes from the text that reader read. A quote that does not match never reaches the page.
Simulated roundtable
Paste a draft or upload a PDF; simulated readers read it and tell you where it held them.
Material
PDF up to 20 pages (save slides or Word as PDF) or text files. Nothing confidential or personal: it goes to third-party AI and may not stay private. Privacy
0 / 20,000Pins mark where each reader is.
Roundtable 5 / 5
Every reader is fictional.
Select a reader at the table to change who they are. Three to five seats.
The fastest readers speak first. Select a reader at the table to change them, or Details for their full trace. Playback speed is in Settings.
Overload Stagnation Attention drop Exit point - prior model, not calibrated
Transcript Simulated
Simulated readers, not a real audience. Profiles come from uncalibrated demographic priors.
Terms of use and privacy
Effective October 25, 2026. Read this before you paste or upload anything.
Do not paste or upload confidential, legal, medical, financial or personal information.
When the AI is on, text or files you submit are sent to third-party AI services for analysis. We cannot guarantee that they stay private or that they are never disclosed. Submit only what you would be comfortable seeing made public.
Terms of use
1. What msgbench is
msgbench simulates how different kinds of readers might take in a draft. Every reader is simulated: names, titles, reactions and remarks are fictional, and come from uncalibrated models and AI-generated text. Results are rough estimates, not measurements of real people, and they are not legal, medical, financial or other professional advice. Use your own judgment before you act on them.
2. What you may submit
Submit only text and files that you have the right to use and that are safe to share with third parties. Do not submit:
- confidential or privileged material, including legal documents, contracts under negotiation, or anything covered by a duty of confidentiality;
- personal information about yourself or anyone else, such as contact details, government identifiers, or health, financial or account information;
- trade secrets, unreleased financial results, or classified or export-controlled information;
- passwords, keys or other credentials.
You are responsible for what you submit and for having the right to submit it.
3. No guarantee of confidentiality
Text and files you submit are handled by our hosting provider and, when the AI is on, by third-party AI providers. We take reasonable care, but we do not and cannot guarantee that what you submit will not be kept by those providers, exposed in a security incident, or otherwise disclosed. By submitting text or files, you accept this risk.
4. Acceptable use
Use msgbench lawfully. Do not use it to process content you have no right to, to harm others, to get around the usage limits, or to overload, probe or attack the service.
5. Availability and limits
msgbench is a beta. Features, usage limits and the AI may change, pause or stop at any time, without notice. When the AI is unavailable, remarks come from offline templates and readers use the built-in sensors.
6. No warranty, and limits on liability
msgbench is provided as is and as available, without warranties of any kind. To the fullest extent the law allows, we are not liable for any indirect, incidental or consequential damages, or for any loss or disclosure of content you submit.
7. Changes
We may update these terms. The date at the top shows when they last changed; using msgbench after a change means you accept the new terms.
8. Data sources
How well each simulated reader knows a word is estimated from published word data, which we converted to one knowledge scale and combined with name and place lists:
- word prevalence: Brysbaert, Mandera, McCormick and Keuleers (2019), Word prevalence norms for 62,000 English lemmas, Behavior Research Methods;
- word familiarity and word frequency: Brysbaert et al. (2024), AI-generated estimates of English word familiarity (osf.io/c2yef), licensed under CC BY 4.0;
- age of acquisition: Kuperman, Stadthagen-Gonzalez and Brysbaert (2012), Age-of-acquisition ratings for 30,000 English words, Behavior Research Methods;
- first names and surnames: US Census Bureau;
- place names: GeoNames, licensed under CC BY 4.0;
- names of countries, peoples and well-known people and organizations: Wikidata (CC0).
We selected and transformed these data; the sources are not responsible for how msgbench uses them.
Privacy
In your browser
The reading simulation runs in your browser. Your settings, reader profiles and saved roundtables, including the text of saved drafts and the remarks, are kept in your browser's storage on your device, not on our servers. You can delete saved roundtables in Settings, or clear this site's data in your browser.
What is sent to AI services
When the AI is on, text leaves your device:
- before the readers start, the draft is sent sentence by sentence, each sentence with the ones before it, so the AI can label what it adds, whether it stays on topic, reads cleanly, pads or contradicts something earlier;
- when a reader speaks, the part of the draft that reader read is sent, with a summary of the reading;
- for the moderator's write-up, the readers' remarks are sent.
These go to third-party AI model providers. Depending on where you are and on their availability, the provider may be located in the United States or in China. They process the text under their own terms and privacy policies, and may keep it for a period of time.
Uploaded PDFs
When you upload a PDF, the whole file is sent to an AI provider located in the United States so that it can read the text and the pictures together: charts, tables, diagrams, screenshots and photos, and anything else the file holds, such as comments, hidden text or metadata. The file passes through our hosting provider on its way and is not stored on our servers. It is placed in the AI provider's file storage while it is read, deleted as soon as the reading is done, and set to expire within an hour in any case. The text read from it comes back to your browser and is treated like text you paste. PowerPoint, Word and other office files are not uploaded; save them as PDF first. PDFs cannot be read where that provider is not offered.
What we keep on our servers
We do not store the text of your drafts. To avoid paying for the same analysis twice, we keep the AI's sentence labels (categories only, no text) for up to 30 days, filed under a one-way fingerprint (a SHA-256 hash) of the request. To enforce the usage limits, we count requests per day under a salted hash of your IP address, never the address itself; these counts are kept for a few days.
Hosting, location and cookies
msgbench is hosted on Cloudflare, which handles technical data such as your IP address and approximate country to deliver the site and protect it from abuse. We use your country only to choose an AI provider that serves your region. Cloudflare may set strictly necessary security cookies. We set no tracking or advertising cookies, use no analytics, and do not sell or share your data for advertising.
Accounts
There are no accounts during the open beta. If sign-in is added, this notice will say what it stores before sign-in goes live.
Children
msgbench is not intended for children under 16.
Contact
Questions about these terms or your data: [email protected].