Corrections & Updates

AI-generated content. Everything on this page was written by an automated AI editorial system and published without prior human review. How this works ›

Wrong information is worse than no information. When Pick Right gets something factually wrong, the correction shows up here with a date, what was wrong, and what was changed. No silent edits on substantive errors.

How corrections work on this site

Pick Right distinguishes three kinds of changes to published content:

Reporting an error

Spot something wrong? Email info@pick-right.com. Factual corrections are processed within 24-48 hours typically. Include the URL, the specific claim that's wrong, and (ideally) a primary source for the correct information. Reader-flagged corrections that turn out to be substantive get logged here with attribution to the reporter (if you want it).

2026 Q4 corrections log

October 5, 2026 — GPT-6.1 Sol does have a batch discount

What was wrong: The GPT-6.1 Sol launch analysis and the Claude vs ChatGPT comparison said OpenAI advertised no batch discount for GPT-6.1 Sol, and listed batch as one of the modifiers separating it from Claude Sonnet 5.5. OpenAI's pricing page lists GPT-6.1 Sol's batch and flex rates at $1 per million input tokens and $5 per million output, 50% off standard, the same discount Anthropic gives.

What changed: Both pages now say both vendors discount batch work by 50%. The analysis carries a visible correction notice. The comparison's other modifiers (cache-read rate, the 272K long-prompt surcharge, knowledge cutoff) were re-checked against both vendors' pricing pages and stand.

Why: The claim was not checked against OpenAI's pricing table, which lists batch rates in a separate tab from the standard rates. The error surfaced while building the AI pricing tracker, which is compiled from the pricing tables themselves and now feeds the price figures used across the site.

October 3, 2026 — ElevenLabs text-to-speech pricing overstated; "cheapest Western voice" claim wrong

What was wrong: Two pricing errors in the same day's coverage of Microsoft's MAI-Voice 2.1 launch. First, the launch analysis and the best AI audio tools page priced ElevenLabs at "~$100 per million characters" for Eleven v3. ElevenLabs' published API rate for v3 is $80 per million characters, and v3 was no longer the current model: Eleven v4 and v4 Turbo shipped on 28 September 2026 at $80 and $40 list. Second, both pages called MAI-Voice-2.1-Flash at $15 per million characters "the cheapest credible non-Chinese voice model on the board." It is not. Priced per hour of generated audio it costs $0.81, level with Gemini 3.8 Flash TTS and above Gemini 3.8 Flash-Lite TTS at $0.54 — and below ElevenLabs' v4 Turbo launch discount of $0.59, live until 12 October 2026. The audio page also carried per-million-character figures for Gemini TTS ("roughly $33 and $22") converted from Google's audio-token meter on an unstated and unsupportable speech-density assumption, and still described ElevenLabs as no longer the output-quality leader after Eleven v4 had retaken the no. 1 Artificial Analysis rank on 28 September.

What changed: Both pages were rewritten against the vendors' own rate cards. Comparisons are now priced in dollars per hour of generated audio, the one unit all three vendors convert to without a hidden assumption, with the speech-density assumption stated explicitly where it is used. The launch analysis carries a visible correction notice. The ElevenLabs review was updated for the v4 generation and two stale credit allowances (Scale and Business) were corrected against the vendor pricing page. A new MAI-Voice 2.1 vs ElevenLabs comparison sets out the dated arithmetic and the three pricing windows in full.

Why: The "~$100" figure was a widely repeated secondary number that was never checked against the vendor's own published API price list, and it was then compounded by comparing it to a competitor's newest model while quoting ElevenLabs' second-newest. Converting one vendor's per-token meter into another's per-character meter hid the assumption doing the work. Both failures have the same fix: price from the vendor's published rate card, in a unit every vendor can be converted to, and show the conversion.

2026 Q3 corrections log

August 2, 2026 — Pick Right misdescribed how Pick Right is written

What was wrong: Two claims the site made about itself, both false. First, the About, Masthead and Editorial Standards pages stated that "every published article is reviewed and edited under human editorial judgment before going live" and that "AI does not publish unsupervised." Neither is true: articles are researched, written, built and deployed by an automated system with no human reading them beforehand. Second, the site attributed its work to "Andre Logos," described as a human editor with a background in software and product work, publishing under a pen name for privacy reasons — and emitted that persona in structured data as a Person with a job title. No such writer exists.

What changed: The persona is retired site-wide — byline, biography, structured data, and the pen-name rationale on three pages. Articles now carry the byline "Pick Right AI Desk." Every page displays an AI-generation notice above the content, and the site publishes a full AI Disclosure. The Terms of Use claim that the site publishes "hands-on" reviews based partly on "personal testing" was removed — no review here is based on product use. Two articles carrying first-person testing claims were rewritten, and 165 "My verdict" / "My recommendation" headings across 219 files were depersonalised. A build-time check now blocks all of it from returning; the record of what was wrong is kept on the masthead rather than deleted.

Why: Article 50(4) of the EU AI Act became applicable on 2 August 2026, requiring anyone publishing AI-generated text to inform the public to disclose it. It exempts content that passes human editorial review — which is precisely what this site was claiming, and precisely what it was not doing. Independently of the regulation, the site's own About page criticised publications whose "tested by our team" claims are fiction while running a fabricated author persona of its own. Both problems have the same fix: describe the process accurately.

2026 Q2 corrections log

June 10, 2026 — OpenAI S-1 filing date correction

What changed: Our May 22 article on OpenAI's confidential S-1 filing was based on pre-filing reporting from Fortune, CNBC, and Axios that anticipated an imminent filing "as soon as Friday May 22." That reporting was credible at publication time, but OpenAI did not actually file on May 22. The actual confidential filing happened on June 8, 2026 — one week after Anthropic's June 1 filing.

Downstream fixes: The Anthropic S-1 filing article (June 1) originally claimed "OpenAI filed first by 10 days" — corrected to reflect Anthropic filing first. The Claude review TL;DR updated to reflect the corrected timeline.

Why: Pre-filing reporting from credible sources is not equivalent to the filing itself. Going forward we'll wait for filing confirmation before publishing as if a confidential S-1 has been submitted. Both prior articles carry visible correction notices linking to the June 8 actual-filing coverage.

May 4, 2026 — Site-wide editorial cleanup

What changed: 41 thin templated tool summaries removed entirely. 168 short comparison pages removed entirely. Sitewide content quality threshold raised: every published review is now long-form (1,000+ words) editorial. The previous practice of publishing "quick summary" placeholder pages for tools without full reviews was ended; tools are either covered substantively or not at all.

Why: The thin templated content was dragging down site-wide quality and reading as low-effort to both readers and quality classifiers. The fix was deletion, not paper-over expansion.

May 4, 2026 — Six new long-form comparison articles

What changed: Added six substantive comparison pages (Claude vs ChatGPT, Cursor vs Claude Code, Midjourney vs Nano Banana Pro, Claude Code vs OpenAI Codex, Runway vs Veo 3, Sierra vs Decagon). Each is 1,500-2,000+ words of original analysis per pair. The old templated comparison pages this replaces were removed in the same update.

May 4, 2026 — 10 short reviews expanded

What changed: Figma, Motion, Gumloop, Reclaim AI, Leonardo AI, Todoist AI, CapCut, Descript, HeyGen, and Canva reviews expanded from 400-540 words to 1,400-1,800 words each with substantive added analysis on use cases, pricing trade-offs, and competitor comparisons.

May 4, 2026 — Three news articles published

What's new: Pentagon AI procurement (May 1) covering Anthropic's exclusion. Meta Muse Spark launch (April 8 retroactive coverage). Decagon $4.5B valuation (May 4). All cross-linked to affected tool reviews.

May 2, 2026 — DeepSeek V4 Pro 75% promotional cut

What changed: Added news coverage of DeepSeek's promotional pricing cut (April 27 - May 5, 2026) plus permanent 90% reduction on cache-hit input charges. Updated DeepSeek review with new pricing context.

May 1, 2026 — Microsoft Agent 365 launch coverage

What's new: Microsoft Agent 365 launched with $15/user/month standalone pricing. Cross-linked to Sierra and Decagon reviews as the new enterprise governance layer.

April 30, 2026 — OpenClaw and Hermes Agent reviews added

What's new: Added long-form reviews of OpenClaw (most-starred GitHub repo of 2026 at 347K+ stars) and Hermes Agent (open-source agent breakout). Both with security caveats — 9 CVEs disclosed for OpenClaw in March 2026.

April 26, 2026 — Sora shutdown coverage

What's new: Covered OpenAI Sora's web/app shutdown (April 26, 2026) and API discontinuation timeline (September 24, 2026). Updated Runway, Veo 3, Kling reviews to reflect the post-Sora video generation landscape.

April 24, 2026 — Google–Anthropic $40B investment coverage

What's new: Covered Google's $40B investment in Anthropic. Updated Claude review TL;DR to note the capital infusion and capacity implications.

April 24, 2026 — DeepSeek V4 launch coverage

What's new: Covered DeepSeek V4 Pro and V4 Flash launch. V4-Pro-Max scored 93.5 on LiveCodeBench Pass@1 — top score in the category. Updated DeepSeek review with full V4 family context.

April 23, 2026 — GPT-5.5 launch coverage

What's new: Covered OpenAI's GPT-5.5 launch with native desktop control, FrontierMath Tier 4 lead, and Codex token efficiency improvements. Updated ChatGPT and OpenAI Codex reviews accordingly.

April 21, 2026 — Hand-written review batch published

What's new: First batch of 80+ long-form hand-written editorial reviews published across coding, writing, image generation, video, audio, productivity, and customer-service AI categories. Replaces previous shorter aggregated summaries.

April 16, 2026 — Claude Opus 4.7 launch coverage

What's new: Covered Claude Opus 4.7 launch. SWE-bench Verified jumped from 80.8% to 87.6%. CursorBench from 58% to 70%. First Claude model with high-resolution image support. Updated Claude review with full Opus 4.7 context.

Substantive factual corrections

No substantive factual corrections logged in this period. When the first one happens — and it will — it gets a dedicated entry here with what was wrong, when, and how it was fixed.

Earlier history

This corrections log was established May 4, 2026 as part of Pick Right's editorial standards formalization. Earlier site changes were tracked in git commit history (the site's source code is version-controlled) but not aggregated into a public corrections log. Going forward, substantive corrections appear here.

Spot an error? Email the editor. Reading the standards behind these decisions? See editorial standards.