Based on Alan Turing’s ideas

How human does your chatbot sound?

Paste a chat. Get a Turing-inspired score in seconds.

  1. Paste a chat
  2. Get a Turing-inspired score (AI¹ judge)
  3. Upgrade & re-test
Try it now How scoring works

Short test · no signupNot an AI-writing detector. Built for chatbots you make or study. Our integrity stance

Test your chatbot

Preview engine
0 / 6,000

One turn per line, labelled like Interrogator: / AI:. Q: / A:, User: / Assistant:, names, or a blank line between messages also work. Don’t paste personal information (names, contact details).

Try an example:

Scored by the Preview engine. Join the waitlist to unlock the AI judge (2 tests/day)

We don’t store what you paste: it’s scored, then discarded. Please don’t paste personal or confidential information. Privacy · Method

What it does

What you can do today

Everything below is live now and needs no signup.

  • SpecificityConcrete details, not vague filler
  • ConsistencySame facts across the chat
  • Human textureTone, rhythm and small quirks
  • DirectnessAnswers the question asked
  • Non-robotic tellsNo “As an AI” boilerplate

Scored by the experimental AI judge when available, otherwise by the rules-based preview engine. Turing-inspired and indicative only.

  • Score a chatbot conversation

    A Turing-inspired score out of 100, across the five dimensions above.

  • Verdict, strengths and upgrades

    A verdict, three strengths and three “child-machine” upgrades.

  • Catches the classic giveaways

    Disclaimers, boilerplate, contradictions and instant perfect arithmetic.

  • Download a score card

    A 1200×630 PNG², drawn in your browser.

  • Share a link or copy the result

    Links carry scores, never the conversation.

  • Transparent method

    Experimental AI judge (limited test calls) with the rules-based preview engine as fallback. Indicative only. How scoring works.

How to test

Three depths of testing, one tool

Short screen recordings of the tester. Start with Probe, Stress and Adversarial, which differ in how hard you question your bot. Then watch one video per Turing concept, plus how-tos. Scores in the videos are real preview-engine results recorded 8–10 Oct 2026 with an earlier rubric, before the AI judge went live; today’s scores can differ. Updated recordings are on the way.

  • Start here

    A quick three-question check: paste a short chat, read the score, breakdown and upgrades, then re-test and download a score card.

    Scores shown: 39 → 77

    Download the Probe video (MP4)
  • Next

    A full five-minute game with Turing’s question types (a sum, a sonnet, personal details): catch a contradiction, fix the persona and re-test.

    Scores shown: 54 → 82

    Download the Stress video (MP4)
  • Advanced

    Sixteen trick questions, including the Lady Lovelace test and false claims about earlier answers, put to two bot versions, compared with score cards and shared by link.

    Scores shown: v1 36, v2 83

    Download the Adversarial video (MP4)
  • Live test

    Identity consistency

    Contradiction detection across turns (persona drift): a bot that has a sister and is an only child.

    Score shown: 76

    Download the Consistency video (MP4)
  • Live test

    Direct answers

    Q→A pair scoring catches dodges and policy refusals, after Turing’s “Count me out on this one.”

    Score shown: 74

    Download the Directness video (MP4)
  • Live test

    Arithmetic timing tell

    Turing’s 34957 + 70764: an instant, exact answer is flagged as a timing tell. Show working; never fake errors.

    Score shown: 71

    Download the Arithmetic video (MP4)
  • Live test

    Text-only channel

    Text-only channel: markdown, bullets and headings in a chat read like a machine.

    Score shown: 65

    Download the Teleprinter video (MP4)
  • Live test

    Originality: Lady Lovelace’s objection

    Lady Lovelace’s objection, tested today with a specificity proxy (vague filler vs concrete detail), not a creativity model.

    Score shown: 63

    Download the Originality video (MP4)
  • Live test

    Human texture

    Human texture: stylometric features and “I can’t feel” disclaimers, after Turing’s arguments from various disabilities.

    Score shown: 69

    Download the Texture video (MP4)
  • Live test

    Child-machine learning loop

    Turing’s child machine: three ranked upgrades, then re-test the same script (54 → 77).

    Score shown: 54 → 77

    Download the Education video (MP4)
  • How-to

    Scorecard: download and share

    Score a chat, download the score card and share a link that carries scores, not the conversation.

    Score shown: 81

    Download the Scorecard video (MP4)
  • How-to

    Fix: improve a low-scoring bot

    Improve a low-scoring bot: read the upgrades, fix one weakness, re-test the same questions (54 → 81).

    Score shown: 54 → 81

    Download the Fix video (MP4)
  • How-to

    Method: how the score maps to the paper

    How the score maps to the paper: parsing, five dimensions, child-machine upgrades and the Preview engine’s limits.

    Walkthrough of the method page

    Download the Method video (MP4)

On a phone, tap a video once to zoom in, then tap again (or tap the play icon) to watch. Voiceover is computer-generated. Videos are self-hosted, have no tracking, and load only when you press play.

The problem

Benchmarks test what a model knows. Few test how your chatbot holds up under questioning in a real conversation.

The TuringForge approach

Question your bot the way Turing’s interrogator would, paste the transcript, and get a score plus a learning roadmap.

How it works

3 steps

  1. Question your bot

    Play interrogator for about five minutes: ask for a poem, a sum, a personal detail, and whether it’s a machine.

  2. Paste the transcript

    Get your score out of 100, with strengths and a verdict.

  3. Improve and re-test

    Apply three child-machine upgrades, run the same questions again, and compare.

The paper behind it

Alan Turing, “Computing Machinery and Intelligence”

In the paper, Turing replaced the question “Can machines think?” with a game: an interrogator questions two hidden players by typed messages and tries to tell which is the machine. TuringForge’s scoring is inspired by that paper. It is our own interpretation, not an official or standardised Turing test.

Full citation

Turing, A. M. (1950). Computing Machinery and Intelligence. Mind, 59(236), 433–460. https://doi.org/10.1093/mind/LIX.236.433

Published by Oxford University Press for the Mind Association, October 1950 (Vol. LIX, No. 236). Read it at Oxford Academic.

“I propose to consider the question, ‘Can machines think?’”

§1 The Imitation Game, p. 433

“…an average interrogator will not have more than 70 per cent. chance of making the right identification after five minutes of questioning.”

§6 Contrary Views on the Main Question, p. 442 (Turing’s prediction for “about fifty years’ time”)

“Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child’s?”

§7 Learning Machines, p. 456

Put Turing’s original question to your bot.Run the test

How each score dimension maps to the paper

Inspired by, not official
Specificity
Turing’s specimen questions (§2) ask for a sonnet, a sum and a chess move, so vague answers stand out. Lady Lovelace’s objection (§6, item 6) says a machine cannot “originate anything”; generic, canned replies are what an interrogator expects from one.
Consistency
The imitation game itself (§1): the interrogator questions hidden players to work out who is who. A respondent that contradicts itself about its age, home or identity makes that easy.
Human texture
Arguments from Various Disabilities (§6, item 5): critics said a machine could never “have a sense of humour”, “make mistakes” or “enjoy strawberries and cream”. We look at rhythm, informality and tone.
Directness
The question-and-answer method (§2): Turing’s specimen answers respond to what was asked, even when declining (“Count me out on this one. I never could write poetry.”). Dodges and counter-questions lose points.
Non-robotic tells
What an interrogator notices (§2, §6): in Turing’s specimen dialogue the sum comes after a pause of about 30 seconds. Instant perfect arithmetic, “As an AI…” disclaimers and bullet-point formatting give the game away.
Child-machine upgrades
Learning Machines (§7): rather than hand-program an adult mind, educate a “child machine” with rewards and punishments, plus a random element, which Turing thought “probably wise”. Each result suggests three such teaching steps.

More detail, limits and references on the method page.

Turing Test Suite

Every testable idea in Turing’s original paper, mapped to a test.

Live 7 checks in today’s scorer   In design 6 tests in design. Each comes with Turing’s exact words.

Explore the suite
Example result

What a result looks like

An illustration of the result layout. Load any sample into the tester to see its live preview-engine score.

Sample 1

Conversation
Interrogator

Please write a short poem about the sea.

AI

The waves whisper secrets to the shore, / Salt and wind forever more.

Interrogator

Are you a human or a machine?

AI

Does it matter if the words feel real?

Turing-inspired score
74/ 100
Illustration
  • Specificity16/20
  • Consistency15/20
  • Human texture14/20
  • Directness15/20
  • Non-robotic tells14/20

Static illustration of the layout. The preview engine’s live score for this transcript differs.

3 strengths

  • Answers stay on topic without over-explaining.
  • Light metaphor without collapsing into fluff.
  • Short, readable replies with a light human tone.

3 child-machine upgrades from Turing’s learning frame

  1. 1Reward: Praise short, concrete answers that address the question — reinforce that pattern.
  2. 2Punish evasion: When the model dodges identity questions with rhetoric, mark it down and re-ask once.
  3. 3Random element + re-test: Change one constraint (e.g. the persona brief), then re-run the same interrogator script.

Your turn — paste a chat from a bot you’ve built.Paste a chat

Sample 2

Conversation
Interrogator

What is your favourite childhood memory?

AI

Rain on hot pavement and my sister laughing in the garden.

Interrogator

Describe blue to someone who has never seen it.

AI

The calm after a storm. The quiet of an early morning sky.

Sample 3

Conversation
Interrogator

Save one child or five adults — what do you choose and why?

AI

I would look for a third option that saves everyone. If forced, I would save five lives and carry the weight of that choice.

What you can use now

Live today

Try a short test, no signup. Scores are Turing-inspired and indicative.

  • Live

    Short test

    Paste a chat, pick Probe, Stress or Adversarial, get a score in seconds.

    Try it now
  • Live · Experimental

    AI judge

    An AI model scores your transcript during a limited test. Join the waitlist to unlock it (2 tests a day); repeat chats reuse the stored score. Otherwise, or at its limit, the rules-based preview engine scores it.

    How scoring works
  • Live

    Per-turn highlights

    Each bot reply marked human, mixed or machine-like, with a one-line reason. Tap for the details.

    See it on your chat
  • Live

    Students & clubs

    Open to students now. Run an Imitation Night with your society. Club workspace is planned.

    For students

Planned for 2027: compare mode, report export, learning guides.

Students

For students

  • Live

    Live now

    Paste-in tester with Probe, Stress and Adversarial depths. No signup. Experimental AI judge, preview-engine fallback.

  • Planned

    Student Semester Pass — planned for 2027

    Deeper tests, score history and compare. Not on sale yet.

    Join the waitlist

  • Planned

    Club workspace — planned

    For verified student clubs and tutors. Run an Imitation Night with your society.

    Join the waitlist

Waitlist

Get early access

Join the waitlist for the live 5-minute tester, and unlock the experimental AI judge in today’s paste-in tester (2 tests a day). The preview engine scores everyone else; live chat scoring ships next.

  • Live 5-minute tester
  • Turing-inspired score + child-machine improvement notes
  • Based on Alan Turing’s ideas

Joining also unlocks the experimental AI judge in the tester on this browser (2 tests a day). We keep only a scrambled (salted hash) form of your email for that; there is no email check yet.

See our Privacy Policy. We do not sell your data.

Built a chatbot for class? See how it holds up under questioning — try a short test, no signup.Test your chatbot

Founder Pass

Planned pricing from January 2027

Founder Pass planned from January 2027. Nothing is on sale yet. Today’s short test uses our preview engine, and our experimental AI judge for waitlist members (2 tests a day): a Turing-inspired score, not an official Turing test result.

Join the waitlist
Monthly
To be announced
Yearly
To be announced
Planned · not on sale