What GDEV’s Form 6-K actually says

Disclosure. Cifratar.ai is part of the GDEV Inc. portfolio (NASDAQ: GDEV). We are using an affiliated issuer’s own filing as evidence in a market argument, so treat this section as interested testimony and check the numbers against the filing — which is why every one of them is quoted from it below.

GDEV Inc. describes itself in the release as “an international gaming and entertainment company” and reports “its unaudited financial and operational results” for the three and nine months ended 30 September 2025. The report reached the SEC as a Form 6-K — the form a foreign private issuer uses to report interim results — and states that its contents are “incorporated by reference into the Company’s Registration Statement on Form F-3” (SEC Form 6-K, 24 November 2025).

The nine-month table in Exhibit 99.1 reads as follows (SEC Form 6-K, Exhibit 99.1):

For the quarter alone, the release’s own highlights state: “Revenue of $98 million decreased by 12% year-over-year,” “Selling and marketing expenses of $30 million decreased by 43% year-over-year” and “Adjusted EBITDA of $26 million in Q3 2025 increased vs. $17 million in Q3 2024.”

The company gives its own reason for the marketing decline, and it is not an AI story: the Q3 fall in selling and marketing expenses is “driven by our continued focus on improving the efficiency of user acquisition activities,” reflecting “a more selective approach to performance marketing, prioritizing channels that attract players with higher long-term value over broad-scale campaigns aimed at short-term growth.” The words “AI” and “artificial intelligence” do not appear anywhere in the release.

Our arithmetic, not GDEV’s. Over nine months, marketing spend fell by just under a quarter while Adjusted EBITDA nearly doubled, on revenue down 3%. That combination — less marketing, more margin, roughly flat top line — is the shape of a shift from volume buying to selectivity. Adjusted EBITDA is a non-IFRS measure the company defines itself, the figures are unaudited, and the 2024 comparatives were restated; read the direction, not the decimal.

Whether synthetic creators contributed anything to that marketing efficiency is not answerable from this filing, and we are not going to imply it. The issuer attributes the change to user-acquisition discipline. Anyone claiming an AI-creator contribution needs a separate disclosure from the company, and none exists today.

Deliberately absent from this section. The earlier version of this article took its marketing figure from the wrong row of the table above — the game operation cost line, which rose 11% — and built its central claim about a budget cut on top of it. The correct line is selling and marketing at $124 million, down 23%. One row off, in the direction that made the argument louder.

Which public companies are actually documented — and which rows we deleted

These are the rows we can stand behind, each with the document that carries it:

Editorial opinion. A three-row table is a worse table and a better document. The version it replaces had seven rows under the heading “2024–2026”, and not one of the brand rows fell inside that window.

Here is what happened to the deleted rows, because the deletions are the more useful information:

• Calvin Klein (NYSE: PVH) and Samsung. Both activations are real, both are one-off collaborations from the CGI-persona era rather than the AI cycle this article is about, and the only source cited under either row was a marketing blog on our own blacklist. The earlier version also stretched a single Calvin Klein activation into a campaign series and gave the Samsung work a hashtag that does not appear on the page it cited.
• Dior (LVMH) and Noonoouri. The partnership is real, but Noonoouri is a hand-built CGI persona — not an AI character, and publicly on record objecting to that label. The source cited under the row does not mention Dior at all.
• Olaplex (NASDAQ: OLPX) and Aitana López. The cited source does not mention Olaplex. Olaplex has published no statement about a collaboration, and we found no filing, release or company page describing one. Trade press asserts it; the company does not. We do not carry a claim about a public company’s marketing that the company has never made.

Editorial opinion. The pattern in those four deletions is one mechanism, not four mistakes: a URL from a real publication is attached to a claim that the page does not contain. It is the single most common defect in writing about this category, and it is why an article arguing “read the filings” has to hold itself to the same standard first.

The BMW iX2 case: what is documented, and what is not

BMW launched the iX2 with “Make it Real,” a hero film starring Lil Miquela, produced with the agency Monks and directed by Stefanie Soho of BWGTBLD. The agency’s own case page lists the recognition — 1× CLIO, 1× Epica, 3× Lovies — and describes the concept: the character crosses from the virtual world into the physical one and discovers human sensation while driving the car (Monks). The film took Gold at the 2023 Epica Awards in the Online & Viral Films category, which also fixes the campaign’s year (BWGTBLD).

Lil Miquela is a computer-generated character created by Trevor McFedries and Sara DeCou at the Los Angeles startup Brud; the reference record describes her as a virtual singer and social-media personality and nowhere applies an AI label to her (Wikipedia).

No performance metric for this campaign has been published by BMW or by the agency. The agency case page carries awards and craft credits and no impressions, interactions or engagement figures at all.

Editorial opinion. That absence is the finding, and it is more useful to an advertiser than any number would have been. The engagement statistics attached to this campaign in trade coverage — in both directions, flattering and damning — trace back to pages with no attribution and no methodology. We removed every one of them from this article rather than pick the flattering one. The operational consequence: if you are planning a synthetic-creator activation in 2026, there is no public benchmark to plan against. Your own before-and-after measurement, on your own account, is the only instrument that exists, and you should budget for building it.

Editorial opinion. What the case does teach is placement. BMW did not hand a product launch to a virtual character; it inserted one activation into an omnichannel launch and let the character’s existing narrative carry the vehicle. That is how the format is actually used by large advertisers: as one channel in a portfolio, judged on craft, not as a replacement for the brand’s face.

The unit economics of one synthetic creator: Aitana López

Euronews reports that Aitana López, created by the Barcelona agency The Clueless, “can earn up to €10,000 a month, according to her creators, but the average is usually around €3,000,” and that she earns “just over €1,000 per advert.” At the time of the report she had gained “more than 343,000 followers on Instagram” in a year and a half, and had “recently become the face of Big, a sports supplement company” (Euronews, published November 2023, updated 27 December 2024).

Genre note. These are figures the creators gave a journalist, not audited numbers, and the report they come from was last updated in December 2024. They are worth reading as an order of magnitude and nothing finer. Follower counts in particular drift; do not reuse this one as current.

In the same interview, Rubén Cruz — Aitana’s designer and the founder of The Clueless — describes the commercial reason the character exists: the agency’s projects kept stalling for reasons outside its control, and building its own model removed that dependency. He also contrasts that economics with the million-euro rates at the top of the market, which he calls absurd. We are paraphrasing rather than quoting because the direct quotation that circulates in coverage of this project, including in the earlier version of this article, does not appear in the Euronews piece or anywhere else we could find.

Editorial opinion. At those rates Aitana sits in micro-influencer territory, priced comparably to a human creator of similar reach in Spain. The argument for the format at that price is not that it is cheaper per post. It is control: a fixed look across a serial campaign, no scheduling, no personal risk, and the option to run the same character into another language market. Those are operational properties, not a discount, and an advertiser should evaluate them as such.

Editorial opinion. One caution against the standard sales line: a synthetic character does not abolish reputational risk, it relocates it. The risk moves from a person’s private conduct to a production decision — what the character says, who writes it, what it is shown endorsing. The most publicised backlashes in this category have been about creative choices, not about the character having a bad night.

Why an issuer’s disclosure is a better signal — and where that argument stops

Companies listed on Nasdaq and the NYSE disclose material information through a defined set of documents — 10-K, 10-Q, 8-K for domestic issuers; 20-F and 6-K for foreign private issuers — under SEC rules, with earnings-call statements made against a forward-looking-statements disclaimer. An agency press release carries no comparable obligation.

, and a correction to the usual version of this argument. That does not make a filing accurate. The GDEV figures used above are explicitly unaudited; Adjusted EBITDA is a non-IFRS measure the company defines itself and warns “may not be comparable to other similarly titled measures”; the 2024 comparatives in the same table were restated after an error in the treatment of an expired put option liability; and as a foreign private issuer GDEV files a 20-F rather than a 10-K, so quarterly detail arrives by 6-K rather than 10-Q (Exhibit 99.1).

Editorial opinion. The right claim is narrower than “filings are true.” A filing is attributable and correctable on the record: a named officer signs it, it is dated, it is incorporated into registration statements, and errors get restated in public. That is what makes it auditable by a reader. An unattributed engagement statistic on a marketing blog has none of those properties — which is the whole reason this article deleted several of them.

The markers worth watching in consumer-brand filings over 2026–2027: AI-content provenance and deepfake exposure appearing in risk factors; marketing automation named in MD&A as a margin driver rather than an initiative; and AI infrastructure capex broken out at the platform companies. Until at least one of those appears, a brand’s synthetic-creator work is a marketing experiment, not an institutional allocation — and it will not be visible in the numbers.

What the budget data actually shows

The IAB’s Creator Economy 2025: Ad Spend & Strategies report (November 2025) projects rather than reports: “For 2025, we project another 26% of YoY growth from $29.5B to $37.1B, which is about 4x faster than the media industry overall (+5.7%)” for US creator-economy ad spend, with an estimated $43.9B and +18% for 2026 (IAB, November 2025).

On AI inside that spend, the same report finds “Three in four brands are using/planning to use AI for creator marketing tasks,” with 46% currently using it. And on the counter-pressure: “95% of brands have concerns about using AI in creator marketing overall,” the single largest concern being “lack of authenticity or human connection” (base: 453 respondents).

Editorial opinion. Those two findings are usually presented as a contradiction. They are a description of an adoption curve. Three in four are moving; nearly all are uneasy about the same thing while moving. The resolution the market is converging on is a split of jobs rather than a winner: synthetic characters for frequency, coverage and multi-market consistency; human creators for cultural moments and the trust transactions that need a person with lived experience behind them.

Deliberately absent from this section. A widely repeated Gartner prediction — that by 2027 half of influencer marketing budgets will go to authenticity, provenance and anti-deepfake work — is not published here. The prediction may well be sound, but gartner.com is closed to our checks entirely (HTTP 403), and we could not locate a Gartner newsroom release carrying it. Under our own rule, a forecast from a named research firm is publishable only against a document we can open, so it stays out.

What Article 50 of the EU AI Act requires

Article 50 puts obligations on both ends of the pipeline. On producers: “Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated.” On users: “Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake, shall disclose that the content has been artificially generated or manipulated” (EU AI Act, Article 50).

The Article 50 obligations start on 2 August 2026. The penalty regime under Chapter XII has applied since 2 August 2025 — the two dates are frequently merged into one, and they are not the same event (EU AI Act, Article 113).

Editorial opinion. For advertisers the practical reading is short. A strategy whose value depends on the audience not knowing the character is synthetic is on a clock in the EU. A strategy that is open about it is unaffected, and gains a compliance argument it did not have to build. Disclosure is cheap when the character was designed to be disclosed and expensive when it was designed to pass.

Where synthetic creators do not work yet

Editorial opinion. Four boundaries, none of which the current tooling changes.

Mass-reach awareness on a launch. When the objective is very large reach inside a week, established human creators with existing audiences remain the stronger instrument. Synthetic characters at that audience scale are rare enough to be named individually.

High-trust transaction categories. Health, parenting, regulated financial advice — categories where the audience is buying the endorser’s lived experience. Transparency about the character’s nature does not close that gap; it defines it.

Cultural moments. Campaigns built on a person’s story, a public reputation or emotional intimacy are human territory. A synthetic character can extend such a narrative; it cannot originate one.

Regulatory exposure. See Article 50 above. Anything whose mechanism is “indistinguishable from a person” is now a compliance question in Europe rather than a creative one.

A practical watchlist: five documents, not five press releases

1. Meta IR earnings-call transcripts, quarterly. Read Susan Li’s prepared remarks for the automation and creative-tooling lines, and check which quarter a number comes from before repeating it — that is exactly where this article’s predecessor went wrong.
2. GDEV’s Form 6-K, quarterly, and the 20-F annually. Read selling and marketing against Adjusted EBITDA over several periods. One quarter of falling marketing spend is a decision; four is a structure.
3. Risk factors and MD&A in consumer-brand annual reports. The appearance of AI content provenance, synthetic endorsers or deepfake exposure in a 10-K or 20-F is the moment a brand has formalised the channel.
4. The platforms’ own product announcements. Read them literally rather than through coverage. Google’s AI Max, for example, is described by Google as “a new, one-click feature suite — AI Max for Search campaigns” that will “expand upon your existing keywords using broad match and keywordless technology” — a feature set inside campaigns you already run, not a new campaign type and not the end of keywords (Google, 6 May 2025). TikTok positions Symphony Automation as tools to “create TikTok-first ads in seconds” alongside “a single Smart+ performance advertising flow where advertisers can tailor to full automation, partial automation, or fully manual” (TikTok Newsroom, 7 October 2025).
5. Cannes Lions, annually in June. The awards structure encodes where the industry thinks value sits: a Creator & Influencer Marketing section with six subcategories now sits inside the Social & Creator Lions (Cannes Lions).

Editorial opinion. Those five cover nearly all of the primary signal in this category. Everything else is a restatement arriving later, and — as this rewrite demonstrates at some length — frequently a restatement of something the original document never said.

Frequently asked questions

Are synthetic creators replacing human influencers?

No, and the data does not describe a replacement. The IAB’s November 2025 report finds three in four brands using or planning to use AI for creator marketing tasks, and simultaneously that 95% of brands have concerns about using AI in creator marketing overall, with lack of authenticity or human connection the single biggest one. In our editorial view the market is splitting the work rather than substituting it: synthetic characters take frequency, control and multi-market consistency; human creators keep cultural moments and trust transactions.

How much do public companies spend on AI-creators?

There is no such line in any SEC filing, and no issuer breaks it out. What exists is adjacent: on Meta’s Q3 2025 call CFO Susan Li said the annual run-rate of revenue running through Meta’s end-to-end automated solutions had reached $60 billion, and on the Q4 2025 call she said the combined revenue run-rate of video generation tools hit $10 billion in Q4. Both describe automated ad delivery and creative generation, not spend on virtual characters. Anyone quoting them as an AI-creator budget is over-reading the document.

Did GDEV cut its marketing spend because of AI?

The filing does not say so, and neither do we. GDEV’s Form 6-K for the nine months to 30 September 2025 reports revenue of $315 million (−3% year over year), selling and marketing expenses of $124 million (−23%), and Adjusted EBITDA of $64 million (+93%). The company attributes the marketing decline to improved efficiency of user acquisition and a more selective approach to performance marketing; the words “AI” and “artificial intelligence” do not appear in the release. The results are unaudited and Adjusted EBITDA is a non-IFRS measure the company defines itself. Cifratar.ai is part of the GDEV portfolio, so read this section as interested testimony and check the filing.

Are the engagement figures published for AI-creator campaigns trustworthy?

Mostly they should not be used at all. For the best-known case in the category — BMW’s “Make it Real” for the iX2 with Lil Miquela — neither BMW nor the agency has published any performance metric. The agency’s own case page carries awards (1× CLIO, 1× Epica, 3× Lovies) and craft credits, and no impressions, interactions or engagement figures. The statistics circulating for this campaign, in both flattering and critical directions, trace to pages with no attribution or methodology. There is currently no auditable public benchmark for synthetic-creator engagement, which means your own measurement is the only instrument available.

What does an AI-creator actually cost?

The only well-documented data point is Aitana López. Euronews reports, on her creators’ figures, that she can earn up to €10,000 a month with an average around €3,000, and just over €1,000 per advert, at more than 343,000 Instagram followers. Those are creator-reported rather than audited numbers, from a report updated in December 2024, so treat them as an order of magnitude. At that level the pricing sits in micro-influencer territory: the case for the format is control, serial consistency and multi-market reuse, not a per-post discount.

What does EU AI Act Article 50 require, and when?

Article 50 requires providers of AI systems generating synthetic audio, image, video or text to ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated, and requires deployers of systems producing deep-fake image, audio or video content to disclose that the content has been artificially generated or manipulated. The Article 50 obligations start on 2 August 2026; the penalty regime under Chapter XII has applied since 2 August 2025. The two dates are often merged and are not the same event.

Is Lil Miquela a CGI character or an AI one?

CGI, and the distinction matters when writing about this category. Lil Miquela is a computer-generated character created by Trevor McFedries and Sara DeCou at the startup Brud, and the reference record describes her as a virtual singer and social-media personality without applying the AI label. The same applies to Noonoouri, a hand-built CGI persona that has publicly objected to being described as AI. Aitana López, by contrast, is an AI-generated model. Filing all three under one AI label is the most common factual error in coverage of the field.

Where does Cifratar.ai fit?

Cifratar.ai is part of the GDEV Inc. portfolio (NASDAQ: GDEV) and builds the infrastructure layer for synthetic creators — character production plus audience analytics reported to the advertiser. The proposition is not to compete with human creators but to cover the frequency, control and multi-market cases they are structurally bad at, as an added channel inside an existing mix.

Browse the AI-creator catalogue