Errata: Corrections We've Made & How to Report One
See our public errata log: every correction dated, sourced, explained. Learn how to report an error. Run by Alexandra Joly, senior editor.
By Alexandra Joly, Senior Editor · Last verified May 26, 2026 · See our scoring pages and editorial process
This is the errata page. It is the running record of every factual correction we publish on bestgirlfriend.ai, dated, sourced, and explained in five fields per entry. It is the fourth surface of the trust setup: our about page names me as the editor, our scoring pages publish the four frameworks we score against, our editorial process page documents the 12-step workflow, and this page is the log when that workflow produces something I later have to correct.
Most adult-review sites do not publish an errata page. The reason is not that they make fewer mistakes than we do. The reason is that publishing corrections requires admitting publicly that the last version of the page was wrong, and most sites in this category do not want to admit anything publicly. So they edit silently, swap a score from 6 to 8 in the middle of the night, and hope nobody notices. We do not. The audit trail of corrections is itself the trust signal. If you can read what I got wrong six months ago, you can decide whether to trust what I am writing today.
I modeled this page on the corrections logs at Wirecutter, The New York Times, and ProPublica, three consumer publications I read closely in adjacent categories, all three of which publish their errata openly. Adapted to a category where the incumbents have chosen the opposite pattern (silent edits, undated reviews, fabricated bylines, no contact email).
What is the errata page on bestgirlfriend.ai?
The errata page is the public log of every factual correction we publish on bestgirlfriend.ai. Each entry carries the date, the page corrected, what changed, and why. It exists because a publication that never publishes corrections is not error-free, it is opaque, and the audit trail of corrections is itself a trust signal a reader can check.
This page is one of four trust surfaces on the site. Our about page names me and the publisher. Our four scoring pages publish the frameworks (AI companions 8 categories, cam sites 6, adult games 7, real models 6). Our editorial process page documents the 12-step workflow from triage to publish, including the score-lock at Step 7 that fixes the number before any commercial layer touches it. This errata page is the only one of the four expected to start empty and grow over time. An errata page still empty a year after launch would itself be evidence the correction policy is not being applied, not evidence we are perfect.
What counts as a correction worth logging here?
A correction is any change to a published review, pillar, listicle, versus, or guide that alters a factual claim, a sub-score, a composite score, a pricing figure, a CTA destination, or a sourced quotation. Typo fixes and copy edits are not logged. Score changes, dead-link replacements, regulatory updates, ToS-driven retests, and anything that could change a reader decision are.
The line between material and minor matters because logging every typo would dilute the audit trail. The test I apply is the reader-decision test. If a reasonable reader's signup or purchase decision could shift because of the edit, the edit is material and goes on this page. If the same reader would reach the same decision before and after, the edit is minor and lives only in the per-page update log + git history. When I am not sure, I publish the correction here anyway. Over-disclosure is a trust gain; under-disclosure is the failure mode this category is known for.
The seven absolute publish-blocking checks documented at our editorial process page (clickbait, broken or contradicted data, missing affiliate disclosure, on-site explicit images, on-site explicit text, forbidden slug pattern, any reference to a minor) each define a class of failure that, if it ships and gets caught post-publish, triggers a material correction and an entry on this page. The NSFW-04 check (any reference to a minor) is the one with no resolution path, if it ever fires after publish, the page is killed and the kill itself is logged here as a correction.
What does an errata entry actually look like?
Each entry has five fields. The date in ISO 8601 format. The affected page URL. A one-line summary of what changed, in present tense. The reason (reader report, internal audit, retest, regulatory update, source correction). And a cross-link to the update log on the affected page where the same change is mirrored. Material corrections also pin a 60-day notice on top of the affected page.
The five-field structure makes entries diffable, citable, and parseable for AI search engines. A populated entry follows the pattern below. These are template fields, not a real correction.
| Field | Format | Example |
|---|---|---|
| Date | ISO 8601 (YYYY-MM-DD) | 2026-06-14 |
| Page | Canonical URL path | /candy-ai-review |
| Change | One-line summary, present tense | Corrected memory sub-score from 6.0 to 5.0 after retest confirmed 5-7 day ceiling |
| Reason | Reader report, internal audit, retest, regulatory update, source correction | Reader report via editorial@ |
| Update log link | Anchor to the affected page's inline update log block | /candy-ai-review#updatelog |
The same fields populate the per-page update log block at the bottom of every affected review. This page is the cross-page aggregator; the per-page update log is the local detail. Both must agree, and the weekly internal audit checks they do.
How do I report an error?
Email [email protected] with subject line Errata: followed by the page URL. Include the specific claim, the correction you propose, and a sourced citation we can verify. Acknowledgment within two business days. Resolution targets: 48 hours for broken links or contradicted data, 7 days for sub-score disputes, 14 days for full retest requests where the scoring cadence allows it.
The mailbox is [email protected], monitored by me personally, same address used for journalist inquiries and platform contestation. Reader reports, platform operators, journalists, regulators, and security researchers all welcome. Anonymous reports are accepted but a verifiable source citation is required before a correction publishes, I will not change a published claim on the basis of an unverifiable assertion, even one that sounds plausible.
A platform contesting a score follows the same workflow but gets one extra gate: a documented retest under the scoring cadence, never a negotiated score adjustment. Brand emails proposing a score lift in exchange for a higher commission rate, a sponsored placement, an exclusive code, or any other transactional sweetener are logged and ignored. Persistent senders are blocked at the mail-server level. The whole point of our editorial process is that the score is locked before the affiliate catalog is opened, so the commission cannot reach the number. A retest can move the number; a brand email cannot.
How are material corrections distinguished from minor edits?
Material corrections change a factual claim, a sub-score, a composite score, a pricing figure, a CTA destination, or a sourced quotation. Minor edits cover typo fixes, formatting, clarity rewrites that do not change meaning, and Schema.org updates. Material corrections land on this errata page. Minor edits live only in the per-page update log and the git commit history.
The reader-decision test is the call. If a reasonable reader's signup or purchase decision could change because of the edit, it is material and gets logged here. If the same reader would reach the same decision before and after, it is minor and stays in the per-page update log. When the call is close, I default to logging it on this page. The cost of over-disclosure is one extra entry that nobody reads; the cost of under-disclosure is a reader catching a silent edit and losing trust in the whole publication. The first cost is bounded; the second one is not.
Errata log
The log below is the running record of material corrections published on bestgirlfriend.ai. At launch (2026-05-11) the log is empty by definition. The launch baseline below holds the slot until the first real entry replaces it.
- 1
Launch baseline, 2026-05-11
Site goes live with the V1 EN-US plateau. No corrections have been published on any page yet. This placeholder gets replaced by the first material correction when one is logged. Reader reports during the first 30 days post-launch are particularly welcome at
[email protected]; the first month after publish is when most factual issues surface, and that monitoring window is the highest-yield window for catching them.
When the first real entry lands, it will read in the same five-field structure shown in the format table above and will replace this launch baseline. Subsequent entries are appended chronologically, newest first.
Why publish an errata page at all?
Accountability: corrections that are visible can be cited and trusted; corrections that happen silently cannot. Reader contract: a review that says today exactly what it said last month is either still correct or quietly altered; this page removes the ambiguity. Search trust: Google and AI search engines cite sources that show their corrections more confidently than sources that hide them.
The third reason matters for AI search specifically. Citation pipelines for ChatGPT, Perplexity, Claude, and Gemini reward sources that demonstrate an audit trail of corrections, because audit-trail sources are statistically more likely to be factually accurate when sampled across thousands of claims. A page that has never been corrected after publish is harder to cite confidently than a page where the editor has demonstrably been wrong, demonstrably corrected the record on a public surface, and demonstrably continued to test and retest. Honesty about prior errors is not just an editorial value; it is a structural ranking input for the AI-search era.
Most reviewers in this category will not tell you any of that. The category pays well and the bar for credibility is low. They keep the silent-edit path open because the silent-edit path is the path that lets them swap a score from 5 to 8 when a higher commission arrives, with no audit trail to embarrass them later. We do not. The script for an editor who has nothing to hide is exactly this page: every correction dated, every change explained, every reason cited. The script for an editor who has something to hide is no errata page at all.
Related reading
- About bestgirlfriend.ai, Senior Editor profile and bio
- Methodology, the four scoring frameworks in full
- Editorial Process, the 12-step workflow and Score-Lock Framework
- Affiliate Disclosure, the commercial half of the trust contract
Common questions about errata
What is the errata page on bestgirlfriend.ai?
The errata page is the public log of every factual correction we publish on bestgirlfriend.ai. Each entry carries the date, the page corrected, what changed, and why. It exists because a publication that never publishes corrections is not error-free, it is opaque, and the audit trail of corrections is itself a trust signal a reader can check.
What counts as a correction worth logging here?
A correction is any change to a published review, pillar, listicle, versus, or guide that alters a factual claim, a sub-score, a composite score, a pricing figure, a CTA destination, or a sourced quotation. Typo fixes and copy edits are not logged. Score changes, dead-link replacements, regulatory updates, ToS-driven retests, and anything that could change a reader decision are.
What does an errata entry actually look like?
Each entry has five fields. The date in ISO 8601 format. The affected page URL. A one-line summary of what changed, in present tense. The reason (reader report, internal audit, retest, regulatory update, source correction). And a cross-link to the update log on the affected page where the same change is mirrored. Material corrections also pin a 60-day notice on top of the affected page.
How do I report an error to bestgirlfriend.ai?
Email [email protected] with subject line "Errata:" followed by the page URL. Include the specific claim, the correction you propose, and a sourced citation we can verify. Acknowledgment within two business days. Resolution targets: 48 hours for broken links or contradicted data, 7 days for sub-score disputes, 14 days for full retest requests where the scoring cadence allows it.
How are material corrections different from minor edits?
Material corrections change a factual claim, a sub-score, a composite score, a pricing figure, a CTA destination, or a sourced quotation. Minor edits cover typo fixes, formatting, clarity rewrites that do not change meaning, and Schema.org updates. Material corrections land on this errata page. Minor edits live only in the per-page update log and the git commit history.
Where does the errata page fit in the trust setup?
The trust setup has four parts. /about names the editor. /methodology publishes the four scoring pages. /editorial-process documents the 12-step workflow. /errata is the public audit log when the workflow produces an error we later correct. The errata page is the only one of the four expected to start empty and grow monthly.
What corrections have you published so far?
As of May 2026, the errata log is empty because the site is in its launch window. The launch baseline entry above holds the slot until the first material correction lands. The first real entry will replace the placeholder and trigger a newsletter notice to subscribers. We expect the log to grow each month; an editor who claims zero errors after a year of publishing is not a careful editor, just a quiet one.
Why publish an errata page at all?
Accountability: corrections that are visible can be cited and trusted; corrections that happen silently cannot. Reader contract: a review that says today exactly what it said last month is either still correct or quietly altered; this page removes the ambiguity. Search trust: Google and AI search engines cite sources that show their corrections more confidently than sources that hide them.
Does the errata page cover comments or social posts?
No. This page logs corrections to on-site editorial content only. Social posts, newsletter issues, and Discord messages have their own correction patterns: a quote-post correction with link to the corrected source on social, a follow-up newsletter for any material correction in a previous issue, and a pinned correction in the relevant Discord channel for community-facing claims.
Do scores ever change after a review is published?
Yes, under the Score-Lock Framework documented at /editorial-process. A composite or sub-score moves only after a documented retest within the published scoring cadence, or after a trigger event (model swap, engine change, ToS update, regulatory action, breach disclosure, UI overhaul). Every score change is logged on this errata page and in the affected page's update log. There is no silent edit path.