Corrections Policy

April 8, 2026 · Cisco Webex and FedRAMP

Our Webex article said Webex is "often the only platform that clears the compliance bar" for government work and that "no other platform matches the government compliance stack". Wrong. Zoom for Government, Microsoft Teams GCC High, and Google Workspace all hold FedRAMP authorizations. The exclusivity claims were removed. The error flattered a vendor and we corrected it unprompted.

April 9, 2026 · Nine unverifiable claims across three articles

A sweep found figures we could not source and removed or softened them: a Fathom user count, an Otter G2 rating movement and a "below 4.0 buyers stop shortlisting" rule, an Otter user and revenue figure, a Gong valuation, named likely acquirers, a Granola public-link exposure written as a confirmed security finding that we could not trace to any investigation, and allegations from an Otter class action stated more firmly than the filing supports. A chart caption was relabelled "rough estimates".

August 2, 2026 · Site claims that were not true

The site layout carried an "All systems operational" indicator that was connected to nothing. It was removed. The About page stat "Tools reviewed" became "Tools cataloged", because most tools have a profile, not a review. The About page's "minimum two-week evaluation" claim and its claim that every review page carries a last-verified date were removed. Both were untrue.

August 11, 2026 · Skribby transcription pricing

We published Skribby's transcription range as $0.39 to $0.70 per hour and described their model list as "10+". Their live pricing lists 30 models topping out at $1.36 per hour. Corrected across the Skribby review, the Skribby vs Recall.ai comparison, the meeting bot API hub, the notetakers directory, and the MeetingBaaS pricing breakdown. Source: skribby.io/pricing.

August 11, 2026 · Skribby price comparison

We said Skribby undercuts Recall.ai by 30%. That compared Skribby's audio-only base rate against Recall.ai's full included feature set. Matched like for like, with video and real-time, Skribby is $0.45 per hour against Recall.ai's $0.50, a 10% gap. Which provider is cheaper all-in depends on the transcription model: Skribby wins with its cheaper models and costs more than Recall.ai with its premium ones.

August 11, 2026 · Skribby storage

We listed "recording storage included" without qualification. Skribby includes one week. Extending to one year costs $0.05 per hour. We have also added the comparison point that Recall.ai's published storage rate covers 30 days, with no one-year rate published.

August 11, 2026 · Recall.ai startup pricing

Our Recall.ai review predated their startup program and did not mention it. Early-stage teams can apply for $0.25 per hour on their first 10,000 hours. We have also noted that volume discounts exist but that the rates are not published. Source: recall.ai/pricing. Submitted by Recall.ai, verified independently.

August 11, 2026 · Recall.ai diarization claim

The Recall.ai profile listed "100% speaker diarization" as a feature and "perfect speaker diarization" as a pro. Neither is a claim we can test. Both now describe the mechanism instead: Recall.ai captures a separate audio stream per participant, so speaker attribution comes from the stream rather than from a model. Logged here on September 7; the change shipped August 11 and was missing from this page until then.

August 11, 2026 · Unsourced Recall.ai estimates removed

Our MeetingBaaS pricing breakdown carried an estimated monthly cost for Recall.ai's volume-discounted Launch tier, and claimed those tiers become the most competitive option at scale. Recall.ai does not publish volume rates, so neither claim could be sourced. Both are gone.

August 11, 2026 · Nylas monthly minimum

We listed Nylas Notetaker at $0.70 per hour without noting the $5 per month minimum that includes the first 300 minutes. Added. Source: nylas.com/pricing.

August 17, 2026 · Recall.ai startup program page

When we logged the startup rate on August 11 the only source was a line in Recall.ai's pricing FAQ. Recall.ai has since published a dedicated page for the program, so that is now the source we cite and link. Reading it changed one recommendation: our MeetingBaaS pricing breakdown left Recall.ai out of the under-100-meetings bracket, and at $0.40 per hour all-in it belongs there, ahead of Skribby once video is matched. The page publishes no eligibility criteria, so we say who the rate is for only as far as Recall.ai does.

Updated later the same day: the sentence above was wrong. The page does say who the program is for, in qualitative terms, early-stage, meeting data core to the product, and building a product around recorded conversations. What it sets is no headcount or funding threshold. We have corrected the MeetingBaaS pricing breakdown to match and left the original wording here rather than overwrite it. Source: recall.ai/startups.

September 15, 2026 · Circleback pricing and its free plan

Our Circleback entry listed "Free trial / Starter $16.99 / Pro $29.99" and said in seven places that the product had no permanent free tier. All of it was out of date. Circleback runs four tiers per user per month: a genuine free plan, Pro at $15, Business at $25, and Enterprise at $35. The free plan covers unlimited meetings with 30-day history, in-person capture, and API access. The pricing breakdown, the review, the FAQ, and the cons list were all corrected, and the value grade moved from B+ to A- because the only weakness it cited, the missing free tier, does not exist. The error ran against the vendor and we corrected it unprompted.

September 7, 2026 · This policy, the methodology page, and the Recall.ai profile

Changes made before publishing our Recall.ai interview, all of them about our own pages.

• The "no pre-publication approval" rule above now states its one exception: a named interview subject may check a draft for facts and off-record boundaries. We had promised that to Recall.ai on tape, so the policy had to say so. The methodology page's independence section says the same.
• The methodology page said each tool "gets hands-on testing with real meeting recordings", that we "test actual integrations", that comparisons use "same meeting recordings, same test accounts", that prices are "checked quarterly" and updated "within one business day", and that we "re-verify reviews quarterly". None of that was true. It now says what we do: benchmark transcription accuracy, desk-research the rest, and re-check when a change is flagged.
• The Recall.ai profile called volume discounts "sales-gated". Recall.ai's stated reason for not publishing them is that startup and enterprise discounts differ too much to list; the profile now says that.
• The profile's strengths paragraph and one data field still said "100% accurate" speaker attribution after the August 11 change. Both now describe the mechanism.
• The profile said Recall.ai processes "3+ TB/sec of raw video" and launches "8M+ EC2 instances monthly". We could not find a source for either figure. Replaced with what Recall.ai's engineering blog says: terabytes per second across many thousands of instances. The employee count changed from "~30" to a 30 to 40 estimate, which is what third-party trackers show.
• This page was hidden from search engines and the sitemap until today. It is now linked from the footer.

October 3, 2026 · Fathom does use a bot on Zoom

The Fathom profile, two comparison FAQs, and our post-bot-era article said Fathom joins Zoom through a native integration with no bot. That was wrong. Fathom sends a bot on Zoom, Google Meet, and Microsoft Teams. Since April 2026 its Mac desktop app can also record without a bot, a feature Fathom labels beta, and bot-free video is limited to Zoom. The same profile described the Team plan as billed on top of Premium and including SSO; Team is a standalone plan, and SSO is an Enterprise feature. Sources: fathom.ai/whats-new, fathom.ai/pricing.

October 3, 2026 · Nylas Notetaker pricing

Nylas restructured its pricing on September 9, 2026, and our pages had not caught up. The $5 monthly plan we cited on August 11 is not on the pricing page, and the profile still called the pricing "not transparent". Nylas now publishes four tiers: Free with 5 bot hours, Essentials at $15 a month with 10 hours and $0.80 per extra hour, Pro at $49 a month with 70 hours and $0.70 per extra hour ($43 and $0.65 billed annually), and custom Enterprise. The profile, the meeting bot comparison table, and the MeetingBaaS pricing breakdown now use those figures. Source: nylas.com/pricing.

October 3, 2026 · Recording cost index rebuilt

The April 2026 version of the recording cost index mixed list prices with estimates and carried figures we can no longer support. It listed MeetingBaaS at an estimated $0.40 an hour and AssemblyAI at $0.54; AssemblyAI's pricing page lists $0.15 to $0.21. It also said meeting recording API prices had fallen 40% since 2024, a figure we could not source. The index now covers the seven vendors that publish per-hour meeting bot pricing, priced under four usage profiles from pages checked on October 3, with the raw data published. Transcription-only APIs and consumer notetakers were dropped from it. Source: the rebuilt index.

October 3, 2026 · Correction pass on 16 insight articles

We re-checked every insight article against primary sources. Several described testing, interviews or personal experience that did not happen; this site does desk research, and those passages are gone. Others had facts that were wrong or out of date. Most corrected articles carry a dated note at the top.

• Recall.ai, Up Close: the article said all five open roles on Recall's job board were on-site; the board now lists six, all on-site in San Francisco. We also added that Recall's pay-as-you-go plan includes the first 5 hours of recording free.
• MeetingBaaS Pricing Breakdown: we said Attendee is MIT-licensed and MeetingBaaS's meet-teams-bot uses the Business Source License. Both use the Elastic License 2.0. We corrected Attendee's setup details, removed an unsourced self-hosting cost and an unsourced claim about why Recall.ai cut prices, applied MeetingBaaS's published tier discounts, and added that MeetingBaaS bills for time a bot waits in a lobby.
• Recording Without Bots: we said Google had no equivalent to Zoom's RTMS. Google has the Meet Media API, now a developer preview closed to new signups. We replaced a quote we had attributed to Zoom with Zoom's documented guidance, corrected a claim about the EU AI Act, and updated the Teams, Google Meet and vendor passages.
• Microsoft Teams and third-party recording bots: the article, then titled "Microsoft Teams Is Blocking Third-Party Recording Bots", overstated Microsoft's change. Teams detects external bots and holds them in the lobby for organizer approval by default; blocking is a separate admin policy that is off by default. It also quoted an unnamed ops lead who was not a real interview source, and misdated Teams Premium's AI summaries, which shipped in 2023. We removed the quote, retitled the article and rewrote it around Microsoft's message center posts.
• Granola Review: we said Granola has no Android app and offers no HIPAA coverage. It now has Android and Apple Watch apps, and offers HIPAA with a BAA on its Enterprise plan. We also removed wording that implied we had tested Granola alongside other notetakers; this review is desk research.
• The Hidden Cost of Free Notetakers: we said Fathom does not train on user data (it trains its own models on de-identified data unless you opt out), that Otter keeps deleted data for "legitimate business purposes" (its policy says "as long as necessary"), that Fireflies takes 45 to 90 days to delete data (it says 30), and that Granola's free plan caps meetings (it limits history to 30 days). We also removed a claim that we tested export features.
• Your Meeting Data Is the Next AI Training Goldmine: we said Granola's AI-training opt-out was available only to Enterprise customers. Every Granola user can opt out, and Enterprise workspaces are opted out by default. We also corrected statements about Otter's retention, Granola's HIPAA option and Krisp's on-device processing, and removed an estimated-volume chart and Gong claims we could not source.
• The AI Notetaker Price War: we said Fathom raised a $47M Series B and had about 4 million users. Fathom raised a $17M Series A in 2024, over $30 million in total, and reported over 400,000 monthly active users when Superhuman acquired it on September 14, 2026. We rebuilt a price chart that mixed annual and monthly prices and removed G2 figures we could not date.
• Cisco Webex in 2026: we said the Webex free plan allows 50-minute meetings; it allows 40. We said Teams offers no true end-to-end encryption for group calls; Teams Premium does, for up to 200 participants. We updated Microsoft 365 prices for the July 1, 2026 increase, noted that Microsoft's US website now sells Business Standard only with Copilot ($23.50), and removed a market-share chart that combined incompatible data, a block of administrator quotes, a hardware ranking we could not source, and unsourced claims about Webex usage and ease of use.
• Clari Copilot Pricing: we said the wingman.ai domain redirects to Clari; it now belongs to an unrelated company. We described the Clari and Salesloft merger as a $1.1 billion deal with 5,000+ customers; the companies disclosed no deal value and say they serve more than 4,000 organizations. We removed unsourced headcount, layoff and G2 figures and noted that Clari Copilot now sells as Salesloft Conversation Intelligence.
• The Gong Tax: we said Gong publishes pricing; it does not. We cited conversations with sales leaders that did not take place; those labor figures are now labeled as our assumptions. We fixed three arithmetic errors and re-priced the alternative stack from current vendor pages, which raises its cost from $47,000 to about $81,000 a year and cuts the modeled savings from 69% to about 46%.
• Why Mid-Market Teams Are Leaving Gong: the article said we had watched teams leave Gong "at a dozen companies" and cited conversations with sales teams; neither took place, and the article is desk research. We removed those lines, added Gong's platform fee and updated Fathom, Fireflies and Sybill prices and ratings.
• Meeting Recording Laws by State: the state table listed Oregon and Connecticut as all-party consent states for all communications, cited Delaware's one-party statute as an all-party rule, and listed Michigan and Missouri as one-party states. We rebuilt the table from the Reporters Committee for Freedom of the Press state guides, linked each state, corrected how Google Meet, Zoom, Teams and Gong handle consent, and added the August 13, 2026 ruling in In re Otter.AI Privacy Litigation.
• Meeting Overload: an earlier version was written as a first-person account by an engineering manager who cut their team's meetings. That person and those experiments did not exist. We rewrote it as a summary of published research, removed the invented results and two charts, corrected a Microsoft statistic attributed to the wrong year, and removed two statistics we could not source.
• Voice AI Agents in Meetings: Hype Check: an earlier version said we had tested five voice AI platforms over two weeks. No such testing took place. We removed those claims, rewrote the passages as analysis of vendor documentation, rebuilt a funding chart that understated funding by roughly ten times, and corrected claims about ElevenLabs, Retell's latency and Bland's pricing.
• The Async Video Market Is Consolidating: an earlier version labeled several named, operating companies as at risk or likely to disappear without data to support it, and cited an unsourced figure of 200,000+ companies. We removed those labels and the figure, dated and linked the remaining numbers, noted that mmhmm now operates as Airtime and that Claap is now part of lemlist, and fixed a chart that contradicted the text.

October 4, 2026 · Recall.ai, MeetingBaaS and Nylas profile fixes

While adding sourced verdicts to the three Recall.ai comparison pages we rechecked each vendor's own pages. The Recall.ai profile cited $51M in funding and a $250M valuation. Recall.ai publishes its $38M Series B and $10M Series A but no valuation, so the profile now leads with the Series B and labels the total as the sum of the listed rounds. It also said the Launch tier removes a 500-hour monthly cap, which the pricing page no longer says. We called MeetingBaaS open source; its core is published under the Business Source License 1.1, which is source-available, and we had labelled it bootstrapped without a source. The Nylas profile said speaker names appear on Meet and Teams only; Nylas's documentation describes speaker-labelled transcripts with no platform restriction. Sources: Recall.ai Series B, Recall.ai Series A, MeetingBaaS license, Nylas media formats.

October 4, 2026 · Voice AI agents article rewritten

After the October 3 correction removed invented test results, the voice AI agents article had little left to stand on. We rewrote it from public sources only: which products can speak in a meeting, what Microsoft, Google and Zoom allow, list-price cost per meeting hour, and vendor latency claims next to an independent phone-call benchmark. The funding chart and the speculative "where agents work" grid are gone.