Beautiful.ai

The individual plan on Beautiful.ai's pricing page read $12 a month, billed annually, in the copy this desk saved on 18 August 2026. The same URL, opened in a browser three days later and read in both billing views, showed $14.50. Both copies are kept, both carry a checksum, and neither is the number a buyer pays without committing to a year.

That number is $45, and it is not on the billing toggle at all.

Beautiful.ai is a presentation tool from Beautiful Slides, Inc. of San Francisco. It generates and auto-formats slide decks around a set of constrained layouts the vendor calls Smart Slides. This page is a ledger of what its own published surfaces say about price, cancellation, refunds and output rights, with a date on every line. Nothing here comes from using the product; what this desk does and does not do is written down. Every figure below was read off a document, and every document was kept.

Beautiful.ai pricing: $45 a month with no commitment, $14.50 on a year paid up front

Three numbers matter, and they are easy to confuse because the page does not present them together.

The month-to-month rate is the one this site scores on, because it is the price with no commitment attached and it is the figure that stays comparable across products that bill on different cycles.

The annual figure has a second problem: it did not hold still. On 18 August 2026 the Pro card read "$12" above the words "Billed annually", and the page's own FAQ answer to "How much does Beautiful.ai cost?" agreed, naming "Pro for individual users at $12/month (billed annually)". On 21 August the card read $14.50. I did not watch that change happen and make no claim about why it happened. What this desk holds is two dated readings of one address that disagree, three days apart.

That has a practical consequence. Six other pages about this plan, checked on 21 August 2026, print $12. On 18 August, $12 was the vendor's own figure. If you are pricing this tool from a page that does not say when it was read, you are working from a number that moved once inside a single week.

The Beautiful.ai pricing toggle does not reach the monthly figure

The pricing page carries a "Billed monthly" and "Billed annually" switch and opens on annual. Clicking "Billed monthly" moves the Teams card from $40 to $50 per user. It does not move the Pro card, which stays at $14.50 with "Billed annually" printed beneath it in both states.

The month-to-month Pro rate sits further down the same page, inside a block headed "Need to create a single presentation?" carrying the line "Our monthly option makes one-time projects a breeze. Cancel anytime.", the price "$45", and the caption "For ad hoc projects". Directly above the price are the words "Everything in Pro plan (billed monthly)", so this is the same plan, priced for one month.

A reader reaching for the toggle to see what a month costs will not find it there. That is worth knowing before concluding the plan costs $14.50: the headline figure and the no-commitment figure are roughly three times apart.

What happens if you cancel a Beautiful.ai subscription, and what the two refund answers say

Two of the vendor's own documents answer the refund question, and they do not answer it the same way.

The terms of service, last revised 22 August 2025 and read on 21 August 2026, are unqualified. Section 4.2 closes with "All fees are non-refundable and non-cancellable." Section 6.2 lets a customer end the agreement at any time by emailing support, then adds that "any prepaid fees for the Beautiful.ai Services will not be refunded and any unpaid fees for the remainder of the Subscription Term shall become immediately due and payable". Section 2.1 renews a plan automatically unless written notice arrives before the term expires.

The pricing page's own FAQ, read on 18 August 2026, opens a window the contract never mentions: "Annual subscriptions may be eligible for a partial refund if a request is submitted within 24 hours of the initial purchase or renewal." The same answer adds "Monthly subscriptions are non-refundable." and ends "Please note that all refunds are subject to a processing fee."

Cancelling and getting money back are separate acts here, and only the first is routine. The FAQ puts cancellation on the customer's own account page: "You may cancel your Beautiful.ai subscription on the Billing section of your Profile. When you cancel your subscription, you can use Beautiful.ai Pro until the end of your billing cycle."

The trial deserves reading twice, because the vendor states the mechanism plainly and a skim will still miss it. From the same FAQ: "A free trial requires a credit card to get started, which will be charged in full if you don't cancel your trial before the trial period ends." A second answer repeats it: "After your trial expires, your credit card on file will be charged automatically unless you cancel your Pro subscription before the 14 days are up." The terms fix the trial at fourteen days. There is no free plan on the pricing page, so the trial is the only unpaid route in, and it requires a card number.

Two reviewers describe a different outcome in practice than the contract sets out. A Trustpilot card updated 8 November 2025 reads: "Update: After further discussions, the support team reached out and agreed to issue a refund." A five-star Trustpilot review dated 23 April 2026 reports the same shape of resolution: "They responded to my request with care, showed genuine willingness to help, and approved the refund without any unnecessary complications." Two accounts do not amend an agreement, and nothing here says a refund is owed. What they establish is that asking is not futile.

What a Beautiful.ai subscription buys under the terms of service

The agreement is with Beautiful Slides, Inc., governed by California law with venue in San Francisco County. Three positions in it carry the commercial-terms row on the card below, and each was read off the document rather than off a summary of one.

Indemnity runs partly in the buyer's favour. Section 11.1 opens: "Beautiful.ai will defend at its own expense any claim or action against Customer brought by a third party to the extent that the action is based upon a claim that the Platform and/or Application infringes any copyrights or misappropriates any trade secrets of such third party". A vendor indemnity of any kind was rare across the contracts this desk had read when this card was built. Read the scope before taking comfort from it: the promise covers the platform infringing, not the deck you produce with it. Section 11.2 runs the other way and is wide, reaching claims arising from the customer's own content and from use of the AI features.

Liability is capped low. Section 8 limits each party's total liability to the greater of the fees paid in the preceding twelve months or two hundred dollars.

The vendor discloses that output may repeat. Section 5.2 states that "Output may not be unique to Customer, and the AI Features may generate the same or similar output" for its other customers. That is the vendor's own sentence, not a conclusion drawn from one.

Output rights under a Beautiful.ai subscription: what is yours and what is licensed back

Section 5.2 splits ownership in two, and which side your deck lands on depends on what went into it.

What you brought stays yours. The customer keeps rights to the Customer Content, named in the clause as "the presentations created by Customer via the Service and Output based specifically on Customer Content".

What the vendor supplied does not transfer. "Any Output based on Beautiful.ai Materials shall be deemed" Beautiful.ai Materials in turn, and licensed back to the customer as "a limited, non-exclusive, personal, perpetual, and irrevocable" licence, usable "solely as and to the extent incorporated into such presentations".

The word doing the work there is "personal", and the document never defines it. It could mean non-transferable. It could mean non-commercial. Section 2.2 does not settle the question either, since the licence it grants over the same materials is scoped to "internal business or personal, non-commercial purposes". This card records the position as unclear after a full read, rather than picking the reading that flatters the vendor or the one that alarms the buyer. If a deck built on the supplied template material is going to a client or into a fundraise, that ambiguity belongs in front of whoever advises you rather than inside a review's summary of it.

Beautiful.ai review scores split three ways: 4.6, 2.9 and 3.0

A spread that wide on one product inside one week invites a quick explanation: unhappy customers go to Trustpilot. Before accepting it, read what each platform prints about how its own sample was gathered.

Who invited each Beautiful.ai review sample, in the platforms' own labels

G2's review cards carry labels beside the reviewer. Three of them recur: "Validated Reviewer", "Incentivized", and a source line reading either "Source: G2 invite" or "Source: G2 invite on behalf of seller".

Trustpilot prints a banner on the profile itself: "Asks customers to review", expanded underneath as "This company invites their customers to review, whether positive or negative". The profile has been claimed since January 2021, carries a paid Trustpilot subscription, and the platform records that the company "Replied to 86% of negative reviews".

Both of the large samples are therefore invited samples, which is worth stating because the usual telling of this story treats only one of them as managed. What separates them is the second label: only the G2 cards carry the word "Incentivized". The two records that do not carry it read 2.9 and 3.0.

Product Hunt's nine reviews fall below the smallest sample this site will score, so they feed no row on the card. They are here for what they do to the reading of the other two.

The usability row averages the two large records rather than choosing between them, weighted by sample size: 4.6 across 196 reviews and 2.9 across 242 pool to 73.0 across a combined 438. Both declare the same collection regime, which is the condition for pooling them at all. That single number is an average of a split record, and an average is precisely the shape that hides one.

What a one-star Beautiful.ai review is usually about, and what the five-star ones report

Trustpilot's star distribution on 21 August 2026 runs 50% at one star and 38% at five. The three bands in between hold 7%, 3% and 2%. That is not the curve of a mediocre product. It is two populations.

A read of the 20 most recent reviews on 18 August 2026 shows what separates them, using the absolute publication dates in Trustpilot's own structured data rather than the relative stamps the interface prints. Fourteen were one star and six were five star, with nothing in between. Of the fourteen, twelve concern billing, cancellation or a charge after the trial, and two concern the product itself. Trustpilot's own topic labels name the same cluster, listing "unexpected charges after free" under Subscription and "widespread frustration over unexpected charges" under Payment.

The individual reviews are specific. One dated 11 May 2026 reads: "I cancelled well within the time frame and still got charged." Another, dated 21 July 2026: "1 star because to cancel you have to fill out a form that is not optional."

Of the six five-star reviews in that same sample, five concern the product or the support rather than the bill. One, dated 19 June 2026, reads: "I've used PowerPoint for decades and know it inside and out. But 9 times out of 10 I'll now put a deck together in Beautiful instead."

Charging a card at the end of a trial is common across subscription software, so it would be easy to file this as a category trait and move on. This desk checked instead of assuming: the same recency read was run on another scored product in this category, and there the one-star reviews concentrated on the product rather than on the bill. On that evidence the billing concentration is a reading of this vendor's record, not of the category.

Two records could not be read at all, and the difference between that and finding nothing matters. Capterra returned a bot-verification interstitial that was not bypassed, so this page carries no Capterra figure. Reddit refused three separate methods across six query phrasings on 18 August 2026 and returned no genuine thread, which is a fact about this desk's access rather than about whether the discussion exists.

The one independent discussion that could be opened in full is old. Beautiful.ai launched on Hacker News on 7 June 2018, and the thread argued about whether the product was AI at all: "This does not appear to be AI driven at all (at least from the demo video)." Another commenter took the opposite view in the same thread: "Very impressive demo! As a management consultant who has spent plenty of time in macro-enhanced PowerPoint, I think this looks excellent." Both are eight years old and describe a launch demo, not the 2026 product.

How this Beautiful.ai review reaches 64.2 out of 100

The composite is 64.2 out of 100, computed 21 August 2026, and the band beside it reads below average. Four rows carry it, and the highest of them is not measuring what its position implies.

Value at 40.0 is the row the price drives, and it is the lowest here. It reads the no-commitment rate against a ladder shared by every product this site scores, which is why the annual discount does not lift it: an offer conditioned on a twelve-month lock is not the same offer as one you can stop at the end of a month.

Commercial terms at 64.0 comes from the three clause positions read above. The fourth position that row can hold, what a free tier permits commercially, has nothing to read here because there is no free tier, so it drops out of the average instead of being scored as a failure.

The row that is not a review of the decks: Beautiful.ai's delivery envelope at 90.7

The capability position on this site is reserved for a reading of how good the work is, taken by a party that gains nothing from the result. No such measurement of this product exists. PresentBench is the public board this category reads for whole-deck generation, and its live leaderboard on 18 August 2026 carried no Beautiful.ai row.

What fills the row instead measures the exit, not the artwork. Delivery envelope asks one thing: on a paid plan, what can you take out of Beautiful.ai and keep? Slide quality is outside its reach entirely, so two tools can trade places on this row while the better-looking decks belong to whichever one came second.

Now read 90.7 with its arithmetic showing, because the number is high for a reason that has little to do with the product. The row looks at four things. Two of them have no published answer from the vendor, so they drop out of the average instead of counting against it, and the two that remain are the two this vendor documents best.

Export routes survived, and they are the strong half. The help centre names four of them, including a separate article headed "How do I export as an editable PowerPoint?" that writes out the click path, and a second confirming that "When you export as a PowerPoint, your notes will automatically export as well."

Export fidelity dropped out. Forty-nine help articles under "export" and all three results for "resolution" carry no DPI, no pixel dimension and no vector claim. One near-match was refused rather than counted: a "Max resolution: 1080p" line describes a video uploaded into a deck, not the deck coming out of one.

Watermarking on a paid export dropped out too. Four results for the word include an announcement about guest workspaces and two user comments, and not one is the vendor stating what a paid file carries. An unpublished position is recorded as unpublished, never as a pass. No statement is not a statement of no.

That leaves 90.7 as an average of the questions this vendor answers in public, which makes it the least informative figure on the card. It is high partly because the questions this product would likely score low on had nothing to score. Read it as a description of the exit, not as a verdict on what comes through it.

One honest gap sits inside that row. The second surviving question, whether delivery can be automated rather than driven by hand, is not described on this page: the page it was read from was not kept, and a figure whose source cannot be shown to you is a figure this desk will not narrate. How the rows combine into a composite, and what the bands mean, is set out on the methodology page.

Beautiful.ai alternatives with a score on this site, and what each review measures

Other presentation tools carry a card here, and the useful question about each is what sits in that same capability position, because that is where two cards stop being comparable.

Gamma has a row on PresentBench. Its capability figure is a board reading of generated decks rather than a substitute for one, so it answers the question this card's top row does not.

Decktopus carries the same delivery-envelope criterion in that position, and for the same structural reason: no board has run it either. The readings underneath differ, though. Decktopus's card fills the watermark question from a vendor statement about its free tier, where this card found no published answer at all, so two figures carrying one criterion name are not interchangeable.

Every card in the category, with its rows and dates, sits on the AI presentation tools hub in composite order.

Who should pay $45 for Beautiful.ai, and who should not start the trial

Pay for it if constrained, auto-aligning layouts are the point rather than a limitation, if your decks leave the tool as editable PowerPoint or Google Slides files, and if the monthly cost is small against the hours it removes. On a year paid up front the individual plan read $14.50 a month on 21 August 2026, which is the least expensive way in for anyone confident about twelve months.

Do not enter a card casually. The vendor's own FAQ says the trial charges in full at the end of fourteen days unless you cancel first, the contract says fees are non-refundable and non-cancellable, and the loudest concentration in the public review record is exactly this. Set the reminder before you enter the number, or do not enter it.

Walk away if you need certainty that a deck built on the supplied template material is unambiguously yours to sell or transfer. Section 5.2 does not deliver that certainty, and this page will not manufacture it.

And if your deciding question is how good the slides actually come out, no figure on this card answers it. That measurement does not exist for this product yet, and no row on this card is a stand-in for it, a delivery figure describes the exit, not the deck.

Who writes these pages, and what this desk refuses to claim, is on the about page.