Changelog

A concept page for the query the tracking guide skips past: what counts as AI traffic at all. The operational definition — assistant referrals, AI-search clicks, and the referrer-stripped dark lane — each mapped to where it lands in GA4, with our own July window (109 of 6,156 visits, ChatGPT 82) as the worked example and the ~70% dark share labeled as the third-party estimate it is. AttriFast appears under the site's standing rule for it: disclosed as the founder's product at every mention, demonstrated, never reviewed; its live claims and prices re-checked on attrifast.com Aug 13, 2026.

elevenlabs.io/pricing grants Creator 121,000 credits; elevenlabs.io/pricing/api prints '220,000 characters of TTS (Multilingual v2/v3) included' at the same price — both read Aug 13, 2026. The new page shows the API column is exact arithmetic (price ÷ the $0.10-per-1,000 rate on the same page, at all four tiers) while the credit pool is what the meter enforces, which prices a subscription character at 1.65-1.82× the list rate, worst at Creator. Also caught: the API page's ~220-minute estimate assumes 1,000 characters a minute, brisker than any of our four measured voices.

Both tiers cap at 10 video minutes a month on Synthesia's own grid (read Aug 13, 2026); Starter's upgrade is MP4 download rights and 125+ avatars against Basic's 9. The page also answers 'is Synthesia $18 or $29' — both, on different billing bases, with the page's own stale FAQ as the third wrong answer — and notes the watermark wording present in our Jul 19 read is now absent from the pricing page entirely.

/pricing-history/synthesia and /pricing-history/invideo assemble each vendor's dated reads from this ledger into one timeline. A route exists only where two or more vendor reads are on record — one read is a snapshot, and calling it a history would claim observation we have not done. The route list grows automatically as second reads land. Also this round: the Synthesia affiliate link now deep-links to the price table (Rewardful attributes on any path carrying the referral parameter; script presence on /pricing verified Aug 13), so 'See Synthesia pricing' buttons land where they promise — and HeyGen buttons now say 'Try HeyGen free', because that link lands on the homepage and we could not verify attribution survives a deep link.

The site converted ElevenLabs credits to finished minutes with a single ~835 constant, computed from the first two bench runs. But the published bench data itself contains five renders — the same 991-character script through four premade voices (Rachel 72s → 826 credits/min, Adam 70s → 849, George 65s → 915, Sarah 65s → 915) plus a 3,308-character longform run (234s → 848) — and 835 sits below four of them. Any reader could disprove the point estimate with one render. Every page, calculator and table now carries the measured range, 826-915 by voice, and converts at the five-render mean of 871 where a single number is unavoidable. Creator's headline allowance moves from '≈2.4 hours' to 132-147 minutes. The same sweep fixed a page contradicting our own verified rollover finding.

Re-read synthesia.io/pricing Aug 13, 2026. The plan cards render Starter at $29/mo monthly or $18/mo billed yearly and Creator at $89/mo or $64/mo yearly — the post-cut prices we logged Jul 25. But the FAQ on the same page still answers 'Synthesia costs $29 per month or $264 if you go with an annual plan' and 'Creator costs $89/month or $804/year' — the pre-cut annual totals ($264 = $22/mo, $804 = $67/mo) that the cards stopped showing three weeks ago. One page, two annual prices for each plan — $48 a year apart on Starter, $36 on Creator. Until Synthesia reconciles its own FAQ, the card figures ($216/$768 a year) are the ones its checkout charges — and the third-party roundups quoting $22 or $67 a month are quoting the stale FAQ, not the ladder.

The parity clause said a post naming several tools got a button for every one that resolved or no CTA at all. It was written to stop a comparison steering readers toward whichever side paid — but the version that shipped suppressed the whole CTA whenever a single named tool had no programme, so four paying tools lost their button because of a fifth that had none. Thirteen multi-tool posts were carrying no exit for that reason. The rule now: a button goes to a tool that pays, and a tool that does not pay keeps its section, its rank and its verdict, with an internal link to our own page instead of a free referral. Applied to the video-translation listicle (three buttons), the UGC listicle (three, with Zeely linked to our own review at #4) and the training listicle (two, with Colossyan linked to our pricing breakdown at #2 — its rank is unchanged and the button row skips the middle entry). What did not change is the part that mattered: rankings are written before buttons are attached and are never reordered to put a payer first. The AI-writing and AI-visibility listicles still have no CTA, because not one tool on either page pays us anything.

Our Browse AI credits explainer paired the annual price ($19/month) with the monthly plan's credit allowance (2,000), because Browse AI quotes annual plans in credits per YEAR and monthly plans in credits per MONTH and we read across the two. The real figure at $19/month billed annually is 12,000 credits for the year — roughly 1,000 a month, not 2,000. Anyone who sized a plan on that page budgeted for twice the extraction volume they were buying. Corrected against the pricing page's own plan-selector values on Aug 5, 2026, and the page now publishes both billing bases side by side so the units cannot be crossed again. The Professional figure and the seat counts were right and are unchanged.

On Aug 2 we deleted a lone Apify button from /blog/web-scraping-for-ai-agents because the post is a three-way comparison and Browse AI had no programme to link to; the parity clause says every tool that resolves gets a button or none do. Browse AI's programme came through on Aug 5, so all three are back in the post's own section order — Apify, Firecrawl, Browse AI — and none of them is there because of what it pays. Two notes for whoever maintains the link: verify it with GET, because HEAD answers 404 on partners.browse.ai while the link is healthy, and it lands on the homepage rather than the price table because PartnerStack ignores destination overrides, so the button says "start free" and not "see pricing". Content-only partner like Apify, Firecrawl, Context.dev and VEED: no tools.ts entry and no score, because we have not benched it.

/blog/synthesia-credits-per-minute stated in its methodology that we earn no commission on Synthesia via that page. The page carries a Synthesia affiliate button, so the sentence was false — and false in the direction that matters, on the surface a reader checks to decide how much to trust the numbers. Found while auditing disclosures for the Browse AI launch and corrected the same day. We swept the rest of the site for the same pattern: the two remaining "we earn no commission" lines, both on Runway pages, are accurate — Runway has no programme we are enrolled in, and neither page carries a button.

Every plan card carries a prompt cap and an answer cap — 50/50 on Starter, 100/300 on Basic, 200/600 on Growth — and Writesonic's own docs say the visibility score tracks how often you appear in answers, "not the prompts themselves". Priced on that unit the ladder inverts: Starter is the cheapest tier per tracked prompt ($1.58/mo) and the most expensive per answer ($0.053 vs $0.022 on Basic), 2.4x worse on either billing basis. Three surfaces on writesonic.com disagree about whether Starter tracks one AI platform or three, and the published caps cannot cover 50 prompts across three platforms daily — we publish that as an open question rather than guessing. Also logged: the docs page listing subscription plans was last modified Mar 13, 2026 and now contradicts the live pricing page on articles, audits, users and projects, so do not quote it.

/blog/veed-pricing carries ~100 impressions per 28 days, the most of any page on the site earning nothing. The affiliate link is Cello, not Impact as third-party listings claim, and it lands on veed.io/referral-sign-up rather than the price table — so the button says "Start a VEED account" instead of "See VEED pricing", because the reader will not get a price page. Content-only partner: no tools.ts entry and no score, since we have not benched it.

Firecrawl publishes its whole meter in server-rendered HTML, which is rare in this category, so all of it was read directly on Aug 2, 2026. The angle every roundup misses is that a credit is not a page: scrape, crawl and map bill 1 per page, but search bills 2 per 10 results, interact bills 2 per browser minute, and monitor bills 1 per page per CHECK — so a hundred pages watched hourly is 2,400 credits a day, about half the Hobby allowance. Plans are published yearly-billed only ($0 / $16 / $83 / $333 / $599); the monthly figures are not in fetchable form, and the two integers on the page that look like them are a React reference and a CSS class, so we publish the gap instead of a derived number. Firecrawl is a content-only partner like Apify and Context.dev — no tools.ts entry and no score, because we have not benched it.

/blog/web-scraping-for-ai-agents is titled "Apify vs Firecrawl vs Browse AI" and carried a single Apify CTA. Browse AI has no affiliate programme wired, so it cannot carry a link — and under the parity clause a page presenting several tools as live options gets a button for each or none. A button on only the paying side of a comparison is the vendor behaviour this site exists to be an alternative to, so it is gone. All three can return the day Browse AI is approved.

Eight more posts across Murf, Synthesia, HeyGen and Creatify now carry a CTA, taking blog coverage from 6 to 19. Every figure in the new copy was grepped against the post it sits on first. The remaining 52 are deliberate: the parity clause kills all thirteen multi-tool posts (including the UGC listicle, where three of the four named tools resolve but Zeely does not), rule 3 kills the refund and cancellation pages, and rule 1 kills the sixteen posts about vendors that resolve nowhere. Coverage is no longer the constraint — nine of seventeen tools pay nothing, and docs/affiliate-programmes-to-apply.md records which programmes are worth applying to. Speechify publishes 30% per sale on a 30-day cookie and we run five unpaid Speechify surfaces today.

Four answer pages whose reader is sizing a plan (ElevenLabs credits per minute, HeyGen credits per video, Synthesia cost per minute, the InVideo watermark answer) now carry a CTA after the sources block, plus four ElevenLabs posts and the Murf-vs-ElevenLabs comparison. Every figure in the new copy was checked against the post it sits on. Deliberately none on /blog/elevenlabs-audio-tags or /blog/elevenlabs-adam-voice: both are craft posts whose fix costs nothing and works on the free tier. And a parity rule now has teeth — a post presenting two tools as live options gets a button for each or neither, so /blog/narakeet-vs-elevenlabs gets none despite ElevenLabs paying, because Narakeet resolves nowhere. A button on only the paying side of a comparison is the vendor behaviour this site exists to be an alternative to.

Found by following our own CTA on Aug 2, 2026. The Impact creative's baked-in destination (/make/ai-video-generator-web/) answers HTTP 500, so every click from /blog/invideo-pricing-explained — the site's highest-traffic page — minted a tracking id and then dropped the reader on a broken page. Repointed with Impact's `u=` destination override to invideo.io/pricing, which returns 200 and still stamps irclickid and mpid, so attribution survives.

Tool.commission is the rate a programme offers; it says nothing about whether we are enrolled. The review and scenario templates printed it unconditionally, so /reviews/speechify rendered "Affiliate link · 30% per sale" over an unpaid link to speechify.com — on an anchor the same code correctly marked rel=nofollow, data-aff=0. Same on pictory, basedlabs, runway, jasper, copyai, writesonic, frase and chatgpt. Disclosure now derives from whether the link actually pays: a rate only where one is earned, and "Unpaid link — we earn nothing if you sign up" said plainly otherwise. Three regression tests added. Only six tools are monetised today: ElevenLabs, Murf, Synthesia, HeyGen, Creatify, InVideo.

That sentence rendered on all 18 review pages, and on /how-we-test and /affiliate-disclosure as a site-wide promise, while bench artifacts exist for ElevenLabs and Murf only. It is a first-party experience claim and it was false on 16 pages. It is now gated on bench evidence; unbenched review pages state that the page is verified research rather than a hands-on test, and the two site-wide pages say "where we have run the task" and point at the published coverage count.

All eight calculators ended on a related-questions block with nothing to click — on the highest-intent traffic the site has (the ElevenLabs character-to-minute converter is its best click earner at 7.4% CTR). Six vendor calculators now carry a CTA placed after the result and its caveats, never above them, with the disclosure derived from whether the link pays. Two deliberately keep none: the Narakeet calculator, because no outbound link resolves and the button would 302 readers back to our own index, and the cross-vendor voice cost calculator, because a fixed paid button underneath a page that computes which vendor is cheapest would contradict its own arithmetic.

Amazon, Google Play Books and Apple Books will each narrate a book at no charge, and the terms diverge far more than the price tag suggests. Read from each company's own server-rendered pages Aug 2, 2026: Amazon's Virtual Voice beta contract says "we retain ownership in the AVV Edition file" while capping list price at $3.99-14.99 on a 40% beta royalty; Google charges "no program fee" and lets you download and sell elsewhere, subject to every-country availability and list-price parity; Apple commits that the "Publisher/author retains audiobook rights" with "No restrictions on producing and distributing other versions". Four routes we could NOT read — ACX, Spotify for Authors, Voices by INaudio and Kobo — are published as a gap with the HTTP status each returned, rather than quoted from memory.

Read on heygen.com/terms Aug 2, 2026. Two separate licence sections: paid users "own all rights in your User Input or User Output" including "for commercial purposes", while Free Plan users get only a "limited, non-exclusive, non-transferable, revocable license … solely for personal, non-commercial, and internal evaluation purposes", and that output "may not be sold, sublicensed, redistributed". The clause nobody quotes applies to paid users: on termination for a breach "you must cease use of and delete any Outputs" — on its face reaching videos already published and delivered. HeyGen also pushes the AI-disclosure duty onto the user. Added as a licence section on the pricing explainer rather than a new URL, so one page keeps the intent.

The round-22 retirement swept the rankings and the scenario pages but missed the blog body copy. Eight surviving recommendations were removed, including a tool-fit row on /blog/youtube-ai-voice-monetization — the site's second-highest-impression page — that still pointed marketing-video work at a company in Chapter 7 liquidation. Comparison rows that keep LOVO for reference now carry the retirement in the row itself.

LOVO, INC. filed for Chapter 7 liquidation in the Southern District of New York on May 27, 2026 (case 26-11249), weeks before a hearing in the voice actors' cloning suit against it. We checked lovo.ai, www.lovo.ai and genny.lovo.ai on Aug 1, 2026 — all three answer HTTP 402. LOVO is now marked discontinued, which removes it from every ranking, drops its Review and price markup, and turns all four of its affiliate CTAs into a disabled 'Service discontinued' state. Its review and commercial-rights post now lead with the filing and keep the old licence wording as a record for anyone who already published Genny audio. Every 'pick LOVO for marketing video' recommendation across the scenario pages and alternatives posts is gone.

tools.ts `start` feeds both the JSON-LD Offer.price and the visible verdict card, and three values had gone stale against this site's own research: Frase shipped price "15" while the same page said $39 eighteen times (live: $39 yearly / $49 monthly), Writesonic $20 against a $79 floor, Copy.ai $49 against $24-29 with a free plan that no longer exists. Separately, /ai-video and /ai-writing titled themselves 'tools, tested (2026)' with meta descriptions claiming a bench run, while samples.ts holds artifacts for ElevenLabs and Murf only — both hubs now say 'ranked' and render the coverage count. 21 'on our bench standings' claims in review copy were rewritten, and the footer's 'site-wide re-test sweep' is now a fact-check sweep, which is what actually happens. Three new build gates: metadata may not claim a first-party test without an evidence surface, Offer.price must appear in the rendered page, and retired products may not be named as live.

A dedicated free-tier page. The live free-plan help article was re-read Aug 1, 2026: the current version, vendor-dated Jul 28, no longer contains the word export at all. A four-version timeline quotes the withdrawn wording from dated Internet Archive captures — Nov 15, 2025 still listed '2 Video mins per week', '1 AI credit' and '4 exports/week (with the invideo logo)' — alongside a Free-vs-Plus decision table. Plus pricing stays third-party-reported per our documented rule; the pricing explainer and the watermark answer now cross-link the new page.

Re-checked apify.com/pricing Aug 1, 2026 — every plan price and CU rate from the Jul 12 pass is unchanged. Added the per-plan limit grid (RAM ceilings, concurrency, proxy IPs) and a dated section on the 2026 Actor pricing migration: no new rentals since Apr 1, full retirement into pay-per-usage on Oct 1, pay-per-event now the standard (per Apify's docs and Apr 14, 2026 announcement).

Corrected our stale claim that GA4 ships no AI channel — Google added a default AI Assistant channel on May 13, 2026, before we first published. Added the full-match regex that actually works (the earlier inline pattern would have matched nothing), step-by-step Traffic-acquisition and channel-group instructions, and the two blind spots: Perplexity (unrecognized, still Referral) and AI Overviews clicks (google/organic; Search Console's Jun 3, 2026 Search Generative AI reports show impressions only).

Added an 11-row jurisdiction/platform table checked Aug 1, 2026: Tennessee's ELVIS Act (effective Jul 1, 2024), California AB 2602/AB 1836 and Illinois' replica laws (all effective Jan 1, 2025 — a 2026 date circulating for AB 1836 is wrong), Illinois BIPA, New York § 50-f, the FCC robocall ruling, the still-pending NO FAKES Act (advanced by Senate Judiciary Jun 18, 2026), plus YouTube's own-voice disclosure exemption and TikTok's consent rules.

Added a six-tool index table — entry paid price, meter type and the dated deep-dive post each figure was verified on (Jul 13–22, 2026 passes) — so the hub routes to the per-tool pricing explainers and the voice cost calculator instead of duplicating their ladders. No price was re-derived; each row reuses its deep dive's verified figure.

Re-verified the rollover rule (bank up to 2× quota, hold at most 3×) on ElevenLabs' pricing FAQ and billing docs Aug 1, 2026 — unchanged since the Jul 19 pass. Added the full self-serve ladder read live (Pro $99/600k, Scale $299/1.8M, Business $990/6M — several 2026 roundups still print older, higher figures) and three month-by-month scenarios, computed from the documented rule, showing when a capped balance starts forfeiting credits.

Re-read YouTube's monetization, GenAI-disclosure and AI-labels pages Aug 1, 2026: the July 2026 three-category clarification (effective July 16, per TechRepublic) is unchanged. Added the January 2026 enforcement breakdown (11 terminations plus 5 content wipes, per IBTimes UK), text-to-speech and disclosure FAQs matching how people actually search, and a section showing that two widely circulated rules — blanket synthetic-voice disclosure and a 'warning → 90-day suspension → permanent removal' ladder — appear on no official YouTube page.

One table now carries the numbers this review's queries ask for: the verified plan ladder (Free 10,000 through Business 6,000,000 credits/month), the rollover ceiling (bank up to 2× your quota; balance caps at 3×) and the ~1-credit-per-character rule with our measured ~835 credits per finished minute. Every figure reuses the site's existing dated reads (pricing Jul 1, rollover Jul 19, 2026) — nothing re-estimated.

The direct answer's first sentence now carries both facts searchers ask together: every free-plan export is watermarked, and there is no documented 720p export cap — 720p is InVideo's preview default, per the correction we logged Jul 25. The Studio-vs-AI split and the archived help-centre wording are unchanged.

The page's direct answer and section order now open with the Free/Plus/Max/Generative tiers and the credit mechanics; the free-plan watermark material follows instead of leading, and a free-vs-Plus FAQ was added. No pricing fact changed: paid figures remain third-party-reported pending a live checkout check, and the source snapshot keeps its Jul 23, 2026 verification date.

Re-read on synthesia.io/pricing Jul 25, 2026. The page's own plan data gives Starter $29/mo or $18/mo billed yearly ($216) and Creator $89/mo or $64/mo yearly ($768), under a banner offering plans from $18/month at a 38% saving. Credit pools and stated minutes are unchanged (14,500 credits, up to 120 min; 44,000 credits, up to 360 min). We had published $264 and $804 from the Jul 19 read, which put our cost per finished minute at $2.20 and $2.23; it is $1.80 and $2.13. Corrected in the pricing explainer, the credits calculator, the cost-per-minute answer and the HeyGen head-to-head. Two of our pages had already noticed the discrepancy and hedged rather than resolving it — that hedge is gone.

InVideo's free-plan help article listed four exports a week with the invideo logo as recently as its Nov 2025 version; by the Apr 2026 version every number had become 'limited access', and the current Jun 15, 2026 version does not contain the word watermark at all. The weekly Monday 12 a.m. UTC reset is still stated. We had attributed a four-a-week figure to a Jul 23 pricing-page snapshot in five places, and a ten-a-week figure in a sixth — but invideo.io/pricing serves no plan names, no prices and no watermark text, so that attribution cannot have been right. Every page that quoted a count now names it as withdrawn, and the unsupported 720p free-export cap is gone from the review, the video hub, the watermark answer and the tool's own free-tier label: 720p is InVideo's documented preview default, not an export limit, and the one hands-on third-party report downloaded 1080p. One further correction to our own earlier wording: invideo.io/pricing does serve no plan names or prices, but the vendor does publish Max at $100/month in an FAQ page — well above the ~$60 third parties report for that tier — so third-party InVideo prices are now labelled unconfirmed rather than approximate.

Published three workflow-specific shortlists built from official vendor pages: AI UGC ad generators, AI video translation tools and AI training-video platforms. Each separates research from hands-on testing and includes quick picks, comparison tables, failure modes and a reproducible buyer pilot.

Rebuilt the pricing explainer around the queries appearing in Search Console. InVideo's Jun 2026 help center confirms the no-card free plan, Monday credit reset, paid clean-export rule, no-credit export action and 12-month top-up validity; paid dollar figures remain approximate and clearly labeled.

Re-read on murf.ai/pricing Jul 22, 2026: Murf now lists four plans (Free, Creator, Business, Enterprise). A $49/mo Creator Plus tier and a second Business tier that 2025-era reviews still quote are gone; the top self-serve price is now Business at $66/mo annual ($99 monthly). Creator ($19/mo annual, 24 hrs/yr) and Business's 96 hrs/yr are unchanged — our figures already matched, and we added the explicit month-to-month prices ($29 / $99).

New explainer on CapCut's commercial license: contrary to the common 'Pro gives you a commercial license' claim, CapCut's own Materials License Agreement (updated Jan 2026) gates commercial use per asset, with an all-or-nothing contamination rule. Plus verified Kapwing and Pictory free-tier watermark answers.

Nine more verified answer pages: Fliki commercial-use and free-plan watermark, does-Murf-have-a-watermark (the no-download nuance), Runway free trial, VEED unlimited exports, and cost-intent pages (Synthesia $/minute, Speechify $/hour, Fliki $/video, Opus Clip $/month) that fill content vacuums where no crisp answer ranks.

New pricing/free-plan breakdowns for the AI-writing category — Sudowrite (credits with no fixed word rate), Rytr (usable free tier), Jenni (metered in AI-action counts), Copy.ai (free plan removed / cheap tiers retired), and an 'is Jasper worth it 2026' by-use-case take. Each flags the stale-review traps most write-ups repeat (retired tiers, monthly-vs-annual, 1-credit-=-1-word myths).

New pricing breakdowns for VEED (flags the retired Basic/Business tiers), Fliki (0.5 credits/min meter) and Opus Clip (1 credit = 1 source minute; per-tier minutes flagged as third-party). Plus a new charted 'cheapest AI voice generator' cost-per-finished-minute comparison built on our measured bench data.

New pricing/review coverage for tools none of the programmatic competitors cover: Narakeet (pricing + calculator + two vs pages), NaturalReader (review + pricing), Listnr, Renderforest and Colossyan pricing. Renderforest and Colossyan both recently overhauled their models — the pages flag which widely-cited numbers are now stale, and mark unconfirmed fields (Renderforest free tier / license) rather than guess.

A dated running record of AI voice/video pricing moves this month — Murf's 4-plan consolidation (re-verified), plus scan-flagged Veo/Sora/Kaiber decreases and Jasper/Writesonic/QuillBot plan trims to confirm on vendor pages. Primary-verified and scan-flagged changes are labeled separately.

New 'HeyGen credits, explained' deep-dive (per-model rates, rollover, top-ups) and three new answer pages. Corrected the free-tier detail across the site after re-reading HeyGen's help center: 1-3 videos/month depending on region at 720p (not a flat 3 at 1080p), commercial use barred on free output, and only Business can buy one-off top-up credits.

Read on synthesia.io/pricing Jul 19, 2026: Starter $29/mo or $264/yr (14,500 credits/yr ≈ 120 min), Creator $89/mo or $804/yr (44,000 ≈ 360 min), free Basic 1,200 credits/mo watermarked with no downloads. Per-second deduction confirmed by the page's own 59-second example.

Dedicated pricing explainers now cover every major vendor on the bench: per-unit math, hidden-costs sections, Offer schema, and honest flags where a vendor doesn't publish a rule.

Synthesia rollover (honestly unconfirmed), HeyGen rollover, Murf's single-editor seat wall, ElevenLabs annual-plan mechanics, and the Speechify Premium-vs-Studio split.

Verified on ElevenLabs' pricing FAQ and help center Jul 19, 2026: unused paid-plan credits bank automatically (max balance 3× quota), downgrading or cancelling forfeits them, and prepaid PAYG replaced postpaid overage for new self-serve accounts (12-month expiry).

The exact non-commercial rule, the 'elevenlabs.io' / '11.ai' title-attribution requirement (which now lives in the help center, not the ToS), the $6 commercial floor, and the no-retroactive-license catch.

A new page type for long-tail questions: 40–80 word direct answer first, vendor-page sources, cluster links, QAPage schema. Five seed questions across HeyGen, Speechify, Murf and ElevenLabs.

The calculator matrix grows to five: HeyGen (minutes by model — Avatar IV/V, III, translation) and Speechify Studio (the 1×/3×/30× voiceover/dubbing/avatar spread), plus dedicated Murf and HeyGen pricing breakdowns.

Two more interactive converters join the Runway one: ElevenLabs credits → finished narration/scripts/dubbing (built on our measured ~835 credits/minute), and Synthesia credits → video minutes at the published 2-credits-per-second rate.

Who owns the videos, the stock-avatar paid-promotion ban and what survives cancellation — every clause read on Synthesia's terms, AUP and help center on Jul 19, 2026.

'What went wrong in our test' (v2 tag-text spoken aloud, v3 take nondeterminism, 65–72s pace spread) plus a per-plan cost-per-finished-minute table from the measured ~835 credits/minute figure.

The bench run itself came back artifact-free in one take — logged as such — with the two billing traps (annual toggle, 24h metering) and a cost-per-generated-hour table.

Both products repositioned in 2026 — Writesonic dropped its budget writing tiers (entry now $79/mo), Frase's old $19.99 solo tier is long gone. The comparison now carries the verified plan ladder and a persona matrix.

Interactive credits → seconds-of-video → dollars converter, built from the verified per-model credit rates.

Verified on Synthesia's help center Jul 13, 2026: after a billing period ends you "lose access to Synthesia and all videos in your account," purchases are non-refundable, and accidental cancellations reverse only within 48 hours.

Verified on Runway's help center Jul 13, 2026: unused monthly credits reset on the billing day and don't roll over, while separately purchased credits never expire.

Refund/cancel rules for ElevenLabs, InVideo, Jasper, Pictory, Runway and Synthesia; credits mechanics for Creatify, Speechify Studio, Browse AI and Synthesia; commercial-rights reads for Murf, Speechify and LOVO — each verified on the vendor's own pages and dated.

InVideo's pricing page renders in-browser only, so our figures stay triangulated from third-party sources and are labeled as such — the one money page on this site not read directly from a vendor page.

Verified on Frase's live pricing page Jul 3, 2026: the $19.99 solo tier quoted by still-ranking 2022 reviews no longer exists; current entry is Starter at $39/mo billed yearly. Our review carries the current tiers.

Verified on writesonic.com/pricing Jul 3, 2026: the low-cost individual writing plans no longer appear — new customers start at $79/mo for the AI-search visibility platform. Legacy plan holders should check their account terms.

The identical 991-character brief through Rachel, Adam, George and Sarah — raw MP3s, measured 9.1–12.5s synthesis times, and a 65–72s finished-length spread from casting alone.

Same brief via the official Murf API: 6.3s synthesis, 64 seconds of audio, first take, WAV original downloadable.

All published prices for the voice category were read on the vendors' live pricing pages and dated — the verification cadence every money page now states.