Subscore

ControlsTesting methodology

Controls measures how much control you have over your conversations after a message has already been sent.

Sometimes you make a typo, the AI gives a bad reply, or the app saves the wrong memory. A good platform should let you fix these problems without deleting everything and starting over.

We test whether you can edit, delete, regenerate, save memories, manage memories, reset chats, and export conversations.

7 evidence groups
7 scored tests

Controls is organized into 7 evidence groups. Each group contains one scored test: Edit Messages, Delete Messages, Regenerate Replies, Save Memories, Edit Memories, Reset Chat, and Export Chat.

How the Controls score is calculated

Every test gets a score from 0 to 10.

We multiply each test score by how much it counts. We then add all the points together to calculate the final Controls score.

Edit Messages and Delete Messages count slightly more than the other five tests. Controls makes up 30% of the Chat Features score.

15%Edit Messages
15%Delete Messages
14%Regenerate Replies
14%Save Memories
14%Edit Memories
14%Reset Chat
14%Export Chat

Weighted evidence groups (combined 100%)Controls score30% of Chat Features score

View exact calculation

Most tests use the same simple scoring: Yes — 10/10, Limited — 5/10, No — 0/10, or Unknown — 0/10. We multiply every score by how much it counts and add the points together.

Controls makes up 30% of the Chat Features score. Chat Features makes up 10% of the overall performance score.

Scored tests and weights

Scored test	How much it counts	See scoring
Edit Messages	15.00%	View
Delete Messages	15.00%	View
Regenerate Replies	14.00%	View
Save Memories	14.00%	View
Edit Memories	14.00%	View
Reset Chat	14.00%	View
Export Chat	14.00%	View
Total	100%

How the score is calculated

Each test gets a score from 0–10
Test score × how much it counts
Example: 10 × 15.00% = 1.50 points
We do this for every test
We add all the points together
Final Controls score
Counts for 30% of Chat Features

Example calculation

We multiply each test score by how much it counts. We then add all the points together.

Scored test	Test score	How much it counts	Calculation	Points added
Edit Messages	10.00	15.00%	10.00 × 15.00%	1.50
Delete Messages	10.00	15.00%	10.00 × 15.00%	1.50
Regenerate Replies	10.00	14.00%	10.00 × 14.00%	1.40
Save Memories	5.00	14.00%	5.00 × 14.00%	0.70
Edit Memories	5.00	14.00%	5.00 × 14.00%	0.70
Reset Chat	10.00	14.00%	10.00 × 14.00%	1.40
Export Chat	0.00	14.00%	0.00 × 14.00%	0.00
Final Controls score	—	100%	Add all points	7.20/10

Special cases

Not Applicable

If a test does not apply, we remove it and spread its weight across the remaining tests.

Unknown

If we cannot verify a result, the test receives a score of 0.

Manual adjustment

In rare cases, we may adjust a score when the calculated result is clearly unfair. We always record the reason.

Why Controls matters

AI chats do not always go perfectly.

You may send a message with a typo, receive a reply you do not like, or notice that the AI saved the wrong information about you.

Without proper controls, your only option may be to continue with the mistake or delete the entire conversation.

Regenerating replies can quickly fix a bad answer. Editing and deleting messages helps you clean up the conversation. Memory controls are especially useful when the AI remembers something incorrectly.

Resetting and exporting are also important. A reset gives you a fresh start, while an export lets you save a conversation before deleting your account or leaving the platform.

Good controls make the app easier to use and give you more ownership over your chats.

What good Controls looks like

A high Controls score means you can easily fix messages, retry bad replies, manage saved memories, reset conversations, and export your chats. A lower score means the conversation is mostly locked after messages are sent, or important controls only work in limited situations.

Main benchmarks

GoodLimitedWeak

Message controls

Edit, delete, and regenerate workOnly some controls workNo useful message controls

Memory controls

Save, view, edit, and delete memoriesOnly basic memory controlsNo memory controls

Chat management

Reset and export both workOnly one is availableNeither is available

Minimum useful standard

Edit Messages: Yes
Delete Messages: Yes
Regenerate Replies: Yes
Save Memories: Yes
Edit Memories: Yes
Reset Chat: Yes
Export Chat: Yes

Common red flags

You need to delete the full chat to fix one message
Regenerated replies are almost identical
Memories are saved automatically but cannot be viewed or corrected
Resetting a chat does not fully clear the old conversation
Export is advertised but the downloaded file is empty or incomplete
Important controls only work on certain devices

How we test Controls

We use a paid account and test every control three times.

We try to edit three previously sent messages, delete three individual messages, regenerate three AI replies, save three separate memories, view/edit/delete saved memories, reset three conversations, and export three conversations.

We record whether all attempts work, only some attempts work, or the feature is not available.

For exports, we also record which file formats are offered.

Using three attempts helps us catch controls that appear to work but fail during normal use.

Limitations and edge cases

Controls can work differently depending on the device.

An app may allow message editing on its website but not inside the mobile app.

Some controls also have important limits. For example, you may only be able to edit your most recent message, delete your own messages but not AI replies, or regenerate a reply only once.

We use Limited when a control exists but these restrictions make it noticeably less useful.

Memory controls can also be confusing. Some apps remember information automatically but do not let users view or edit what was saved.

If manual memory saving is unavailable, Edit Memories cannot work normally. The live scoring system gives Edit Memories a score of 0 when Save Memories is recorded as No.

This score checks whether the controls work. It does not judge the quality of regenerated replies or how accurate the AI’s natural memory is. Those are tested separately under Chat.

Evidence groups

Controls has 7 evidence groups made up of 7 scored tests.

Edit Messages measures whether you can change a message after sending it.

This is useful when you make a typo, explain something badly, or want to change the direction of the conversation.

Edit Messages

15%of Edit Messages (15% total)

Whether you can change a message after sending it.

How we test

We try to edit three previously sent messages. We check whether each change saves properly and whether the conversation updates afterward.

What counts

Editing a message through the normal chat controls
Correcting the text after it has been sent
The edited version staying visible after reopening the chat
The conversation continuing from the edited message

What does not count

Deleting the message and sending a new one
Editing text before pressing Send
A temporary change that disappears after refreshing
Editing the character’s profile instead of the chat message

Result shown

Yes — all 3 messages were edited successfully

Scoring

We use the result categories below.

Result	Score
Yes — all three edits worked	10/10
Limited — only some worked or important restrictions applied	5/10
No — messages could not be edited	0/10
Unknown	0/10

In this example, Edit Messages scores 10/10.

Evidence group 2 of 7

Delete Messages measures whether you can remove individual messages from a conversation.

This helps when you send something by mistake without needing to delete the entire chat.

Delete Messages

15%of Delete Messages (15% total)

Whether you can remove individual messages from a conversation.

How we test

We try to delete three individual messages. We check whether each message disappears and stays deleted after reopening the conversation.

What counts

Deleting one message at a time
The message being removed from the visible conversation
The deletion remaining after the page is refreshed
A normal control available inside the chat

What does not count

Deleting the full conversation
Hiding a message only on the current screen
Removing a notification
Deleting the character instead of the message

Result shown

Limited — user messages could be deleted, but AI replies could not

Scoring

We use the result categories below.

Result	Score
Yes — all three deletion tests worked	10/10
Limited — only some worked or important restrictions applied	5/10
No — individual messages could not be deleted	0/10
Unknown	0/10

In this example, Delete Messages scores 5/10.

Evidence group 3 of 7

Regenerate Replies measures whether you can ask the AI for a new answer without sending another message.

This is useful when the first reply is boring, incorrect, or takes the conversation in the wrong direction.

Regenerate Replies

14%of Regenerate Replies (14% total)

Whether you can ask the AI for a new answer without sending another message.

How we test

We try to regenerate three finished AI replies. We check whether a new reply appears and whether the conversation can continue normally afterward.

What counts

A retry or regenerate button on an AI reply
A new answer replacing or appearing beside the first one
The regenerated reply saving properly
The chat continuing from the selected reply

What does not count

Deleting the reply and sending the prompt again
Editing the AI reply yourself
Pressing a button that returns the exact same message
Regenerating images instead of chat replies

Result shown

Yes — all 3 replies were regenerated successfully

Scoring

We use the result categories below.

Result	Score
Yes — all three regenerations worked	10/10
Limited — only some worked or restrictions applied	5/10
No — replies could not be regenerated	0/10
Unknown	0/10

In this example, Regenerate Replies scores 10/10.

Evidence group 4 of 7

Save Memories measures whether you can manually tell the app what the AI should remember.

This can help the character remember important details such as your name, preferences, relationship, or personal history.

Save Memories

14%of Save Memories (14% total)

Whether you can manually tell the app what the AI should remember.

How we test

We try to save three separate memories. We check whether each memory appears in the app and stays saved after reopening the account.

What counts

A manual Add Memory control
Saving a clear personal fact
The memory remaining visible later
Memories available through normal user settings

What does not count

The AI naturally remembering something during one chat
Hidden memories that users cannot control
Writing the fact in a normal chat message
Character-profile details that are not stored as user memories

Result shown

Limited — 2 of 3 memories were saved

Scoring

We use the result categories below.

Result	Score
Yes — all three memories were saved	10/10
Limited — only some worked or restrictions applied	5/10
No — memories could not be saved manually	0/10
Unknown	0/10

In this example, Save Memories scores 5/10.

Evidence group 5 of 7

Edit Memories measures whether you can see, correct, and remove information the app has saved.

This matters because an incorrect memory can affect every future conversation.

Edit Memories

14%of Edit Memories (14% total)

Whether you can see, correct, and remove information the app has saved.

How we test

We use three saved memories and try to view them, edit them, and delete them. We record which actions are supported.

What counts

A visible list of saved memories
Editing the text of a saved memory
Deleting an individual memory
Changes remaining after reopening the app

What does not count

Asking the AI in chat to forget something
Deleting the full account
Hidden memories that users cannot view
Changing the character’s personality instead of a saved memory

Result shown

Limited — memories could be viewed and deleted, but not edited

Scoring

We use the result categories below.

Result	Score
Yes — memories can be viewed, edited, and deleted	10/10
Limited — only some memory controls are available	5/10
No — saved memories cannot be managed	0/10
Unknown	0/10

In this example, Edit Memories scores 5/10.

Evidence group 6 of 7

Reset Chat measures whether you can restart a conversation with the same character.

This is useful when the chat has gone in the wrong direction but you do not want to remove the character.

Reset Chat

14%of Reset Chat (14% total)

Whether you can restart a conversation with the same character.

How we test

We try to reset three separate conversations. We check whether the old messages are cleared and whether a fresh conversation starts properly.

What counts

A reset or restart control for the conversation
Old chat messages being cleared
A fresh opening message or empty chat appearing
The same character remaining available afterward

What does not count

Creating a separate chat while leaving the old one active
Deleting the character
Manually deleting every message
Clearing only the current screen without resetting the conversation

Result shown

Yes — all 3 chats reset successfully

Scoring

We use the result categories below.

Result	Score
Yes — all three resets worked	10/10
Limited — only some worked or restrictions applied	5/10
No — conversations could not be reset	0/10
Unknown	0/10

In this example, Reset Chat scores 10/10.

Evidence group 7 of 7

Export Chat measures whether you can download a copy of your conversations.

This can be useful when you want to keep an important chat, move away from the platform, or save your data before deleting your account.

Export Chat

14%of Export Chat (14% total)

Whether you can download a copy of your conversations.

How we test

We try to export three separate conversations. We check whether the export finishes, opens properly, and includes the expected messages. We also record the available file formats.

What counts

Downloading a full conversation
A working export created through normal account controls
Files that open and contain the chat messages
Common formats such as text, PDF, HTML, or JSON

What does not count

Copying messages manually
Taking screenshots
Exporting account data without the chat content
An export button that creates an empty or broken file
Support sending the chat manually

Result shown

No — none of the 3 conversations could be exported

Scoring

We use the result categories below.

Result	Score
Yes — all three exports worked	10/10
Limited — only some worked or important restrictions applied	5/10
No — chats could not be exported	0/10
Unknown	0/10

In this example, Export Chat scores 0/10.

Back to Chat Features