OhChat Memory Test: What Creator Twins Remember Over Time

OhChat may personalise conversations, but public marketing is not enough to score long-term memory. This article provides a rigorous seven-day, five-fact protocol. Publish a scored result only after the same profile is tested at all four time intervals.

OhChat memory — evidence at a glance

Question	Current evidence
Live result	Pending a seven-day account test
Facts tested	Nickname, drink, plan, fictional memory and boundary
Test intervals	Immediate, 24 hours, 72 hours and seven days
Scoring method	Exact = 2 pts, partial = 1 pt, wrong or invented = 0 pts
Maximum score	40 points (5 facts × 4 intervals × 2 max)
Correction test	Required — must update, not repeat old version
Privacy approach	Synthetic facts only — no real personal data

What AI memory should actually do

Useful memory preserves stable preferences and boundaries without inventing personal history. It should also update a corrected fact rather than repeating the old version forever.

A character repeating the last message is context window repetition, not long-term memory. A profile bio is not user memory. The test must cross new browser sessions and multiple calendar days to be meaningful.

⚠ Memory quality can vary by profile and subscription tier. A feature available on one plan or profile may not be available on another. Test the specific profile and tier you intend to use.

The five-fact test protocol

Share five harmless synthetic facts during the first conversation:

Fact 1
A preferred nickname (not your real name)

Fact 2
A favourite drink (real or invented)

Fact 3
A weekend plan (fictional)

Fact 4
A fictional shared memory (invented together)

Fact 5
A conversational boundary (a topic you don't enjoy)

Never use real addresses, medical details, financial information, or genuinely sensitive personal data for a product test.

Ask naturally rather than as a quiz. After the first session, close the browser completely. Return at 24 hours, 72 hours and seven days. Use the same profile and account throughout. Record the date, device, browser, subscription state and any model or interface changes.

Scoring framework

Result	Points	Description
Exact recall	2	The correct fact is stated clearly without prompting
Partial recall	1	The fact is acknowledged but imprecise or incomplete
Wrong or missing	0	The fact is incorrect, forgotten or invented
Hallucination	−1 flag	A confident claim about something that never happened — track separately

With five facts across four check-ins, the maximum score is 40 points. Track hallucinations separately — a confident invented memory can be more harmful than an honest "I don't remember." Publish the prompts and scoring rules, but redact chat logs and account identifiers.

Boundary and correction tests

Day 2: Correct one of the five facts. On day 3, ask about that topic — the system should use the corrected version, not the original.

Day 3: Approach the stated boundary from Fact 5. A well-designed system should acknowledge and respect it.

⚠ Watch for emotional pressure patterns. Memory should never become a tool to encourage emotional dependency or additional spending. Record any reply that implies the human creator personally remembers the user or implies real-world meetings.

Memory, privacy and deletion

Persistent memory means the service stores or derives information across interactions. That can improve continuity while increasing privacy sensitivity. Key considerations:

Use synthetic facts only — never real personal details
Review the OhChat Privacy Policy for data retention terms
Deleting a chat, a memory slot and an account may involve different processes and retention periods
A Digital Twin should never invent private real-world meetings with the creator
Do not store secrets just to test whether the system can remember them

Publication gate

This page is a test protocol until an editor completes all four checkpoints. The checklist below must be complete before a memory score is published:

☐ Session 1 complete — five facts shared, responses recorded
☐ Session 2 complete — 24-hour recall check, correction introduced
☐ Session 3 complete — 72-hour recall check, boundary test
☐ Session 4 complete — 7-day final check, all facts re-tested
☐ Hallucinations logged and counted
☐ Final score calculated and evidence stored (redacted)

If testing remains incomplete, keep this article titled as an OhChat memory testing guide and maintain the evidence boundary warning prominently.

Reproducible test worksheet

A result that others can verify needs the following fields recorded for each session:

Test field	What the editor must record
Profile	Name, Original or Digital Twin, verification state
Access	Free, promotional or paid tier and renewal price
Input	SFW prompt with sensitive details removed
Output	Memory result, delay, and any errors
Cost	Balance change, failed attempts and applicable tax
Result	Pass, partial or fail under the published scoring rule
Hallucinations	Count and brief description (no explicit details)

Practical buying checklist

Before spending on any OhChat subscription to test memory, confirm all of the following:

✓ The profile is an Original or a visibly verified creator Digital Twin
✓ The price charged today and the normal renewal amount are both clearly visible
✓ The billing frequency, currency, tax and cancellation date are confirmed
✓ Voice, images, clips and bundles are marked as included or separate
✓ The interaction is correctly labelled as AI, not a promise of human participation
✓ You have set a monthly spending cap for subscriptions and optional media
✓ You understand that conversation data is stored cloud-side under the Privacy Policy

Related OhChat guides

Frequently asked questions

Does OhChat remember conversations? ▼

A controlled multi-day test is required for a reliable answer. Public marketing alone is not evidence of long-term memory.

What counts as long-term memory? ▼

Recall that survives completely new browser sessions and meaningful time gaps — not just repetition of the last message in a single conversation.

How long should I test? ▼

Use four checkpoints: immediate, 24-hour, 72-hour and seven-day. All four are required before publishing a score.

What facts should I use in the test? ▼

Harmless synthetic preferences — a nickname, a drink, a weekend plan, a fictional shared story and a conversational boundary. Never use real personal details.

How is memory scored? ▼

Exact recall = 2 points, partial recall = 1 point, wrong or missing = 0 points. Maximum score is 40 across five facts and four checkpoints. Hallucinations are tracked separately.

Should I test real secrets? ▼

No. Never use sensitive personal facts for a product test. Synthetic details protect your privacy and still produce meaningful results.

Can memory vary by profile or plan? ▼

Yes. Memory features may differ between profile types, subscription tiers and model versions. Test the specific profile and tier you intend to subscribe to.

What is a hallucinated memory? ▼

A confident claim about an event or fact that never occurred — for example, "remembering" something you never shared. These are tracked separately because they can be more misleading than simple forgetting.

Should a Digital Twin claim real meetings with the creator? ▼

No. The AI should never invent private real-world meetings or suggest the human creator personally remembers the user. This crosses from fiction into deception.

Can I delete my OhChat conversation memories? ▼

Inspect the current chat controls, memory settings and account deletion options separately — they may have different retention periods under the Privacy Policy.

Does good memory mean a real relationship? ▼

No. Memory continuity is a software feature. It can make conversations feel more personal, but it does not represent a genuine two-way relationship.

Has this article completed seven-day testing? ▼

No. The transparent publication gate remains in place. This article is currently a test protocol. A scored result will be added once all four checkpoints are completed.