How We Test OhChat
Our editorial methodology — what we test, how we score it, and when we wait.
Three article tiers
Our 18 OhChat articles fall into three research tiers, reflecting the evidence available at publication:
| Tier | Description | Publication approach |
|---|---|---|
| Foundation reviews | Pricing, legit checks, free tier, coupons, cancellation — all answerable from public documents and verified company records | Published with desk research evidence and dated claims |
| Experience deep-dives | AI girlfriend, Digital Twins, Originals comparison, uncensored chat, SuperModel profiles, OnlyFans comparison — require platform knowledge but not a live subscriber account | Published with evidence table, clear scoping of what was and was not account-tested |
| Feature tests | Voice messages, image generation, video calls, clips and bundles, memory, character selection — require a real paid account and controlled test protocol | Protocol published first; numeric scores gated until all checkpoints complete |
Desk research standards
For all articles, desk research includes:
- Review of OhChat's public homepage, character directory and subscription pages
- Review of OhChat's Terms of Service and Privacy Policy (source-checked and dated)
- Verification of Utility3 Ltd at UK Companies House (number 14000143)
- Review of OhAPI technical documentation for infrastructure claims
- Cross-referencing public review sources and user reports for consumer experience signals
All desk research claims are linked to their source and include the check date. We do not present inferred or assumed capabilities as verified facts.
Live account testing standards
For feature-test articles, live testing follows this protocol:
- Clean account: A new account is created specifically for testing. No prior subscription history.
- Specified profile: The exact profile tested is recorded (name, type, verification status, tier).
- Defined prompts: Standardised SFW prompts are used so results can be reproduced.
- Controlled scoring: Each feature has a published scoring rubric before testing begins.
- Full sample: We do not stop after the best result. Failures are recorded as part of the cost and quality assessment.
- Dated evidence: Screenshots are taken for every claim, redacted to remove account identifiers, and stored.
- No mixed versions: If the product changes during multi-day testing, the change is noted rather than silently mixing two product states.
Scoring rubrics
Voice messages (5-point framework per message)
| Criterion | What passes |
|---|---|
| Pacing | Natural speed, no rushed or truncated delivery |
| Pronunciation | Common words correctly pronounced |
| Creator likeness | Matches the profile's stated style (Digital Twin only) |
| Delay | Generation time acceptable and recorded |
| Privacy | No unexpected retention of test input beyond session |
Image generation (5-scene SFW protocol)
| Scene type | Scoring criteria |
|---|---|
| Portrait — neutral background | Face consistency, prompt accuracy, artefact count |
| Portrait — styled setting | Background matches prompt, identity preserved |
| Half-body — casual outfit | Proportions, clothing accuracy, face consistency |
| Half-body — described environment | Environment match, identity preserved |
| Full-body — activity pose | Pose accuracy, anatomy, identity preserved |
Each scene is scored 0–5. Results are averaged. Generation time and cost per usable image are recorded.
Memory (7-day, 5-fact protocol)
| Result | Points |
|---|---|
| Exact recall | 2 |
| Partial recall | 1 |
| Wrong or missing | 0 |
| Hallucination (confident false memory) | −1 flag (tracked separately) |
Maximum: 40 points (5 facts × 4 checkpoints × 2). Checkpoints: immediate, 24h, 72h, 7 days.
Publication gates
A feature score is gated until:
- All required test checkpoints are completed
- Evidence (redacted screenshots) is stored
- The scored article has been reviewed against the checklist
During the gate period, articles display a gold "Live test pending" badge and explain the protocol. This is deliberate — publishing a score without evidence would mislead readers and undermine the site's credibility.
What we do not do
- We do not compare unmatched features (e.g. a pre-made clip vs a live video call)
- We do not use API demos as consumer product evidence
- We do not publish explicit content, chat logs or creator likeness imagery
- We do not use real personal data in test protocols
- We do not claim verified scores from incomplete tests
Update schedule
- Pricing articles: Rechecked monthly (prices change frequently)
- Legal, privacy and cancellation: Rechecked quarterly
- Feature tests: Rescored after significant platform updates
- All articles: "Research updated" date revised on any material change