People can no longer tell.
In a peer-reviewed study, people picked out AI-generated faces at a rate below a coin flip. Looking harder does not help.
Verifyco · Pitch · October 2026
A forensic eye for synthetic media.
Know what can actually be shown about a photo, video, public link or document. On iPhone in 20 languages and on the web, honest when the answer is “inconclusive”.
01 · The problem
Seeing is no longer evidence.
Most detectors answer with a black box.
One real-or-fake score rarely shows which checks ran or how certain the system is. On real social-media content, published detectors lose much of their lab performance.
Wrong calls carry real costs.
A wrong “fake” is a false accusation; a wrong “real” is a fraud. Generative-AI-enabled fraud losses in the US are projected to reach US$40 billion by 2027.
Sources: Nightingale & Farid, PNAS 2022 · Reuters Institute Digital News Report 2025 · independent in-the-wild detector benchmarks, 2025 · Deloitte Center for Financial Services, 2024
02 · Mission and thesis
Make the authenticity of digital media reviewable: with evidence, coverage and stated uncertainty, never a black-box verdict.
Signed media is arriving; someone has to read it. Camera makers and AI labs now stamp content with provenance signals, and regulators require labelling. Yet each verifier reads only its own mark. Verifyco brings together the evidence it can read, signed and unsigned, and explains what can be shown.
Evidence, not an oracleEvery result names the checks behind it, what they covered and what did not apply.
Inconclusive is an honest answerWhen the evidence is thin, the answer stays inconclusive and says which way it leans.
On-device analysis on iPhoneFiles you pick on your iPhone are analysed on the phone by default.
People make the final callVerifyco informs a decision. It does not replace the person making it.
03 · Why now
By 2026, provenance became law and hardware.
- TikTok reads Content Credentials on uploads
- Galaxy S25 labels AI edits with Content Credentials
- Pixel 10 signs captures with Content Credentials (C2PA)
- SynthID check ships in the Gemini app
- OpenAI provenance checker; US TAKE IT DOWN Act in force
- EU Code of Practice on AI-content marking, final
- EU AI Act Article 50 transparency duties apply
What Verifyco reads
Content Credentials (C2PA) and, with consent, supported AI-provider signals, plus its own forensic analysis of signed and unsigned media, explained in the user's language.
Who must now act
Platform trust-and-safety teams facing short takedown windows; newsrooms labelling AI content as deployers; fact-checkers working under the EU Digital Services Act election guidelines.
04 · The product · iPhone and web today
One engine for every way media arrives.
Photos & videoHEIC, JPEG, PNG, WebP, GIF, MP4 and MOV. High-resolution photos and 4K video, analysed on the device.
Public linksInstagram, TikTok, YouTube, X and Facebook: paste a post, Verifyco fetches the media and analyses it.
DocumentsPDF structure review: revisions, signatures, active content, embedded files. Diagnostic only.
Forensic reportsVerdict, per-check status and coverage. Full findings and PDF export with Pro.
Share sheetLive ActivityHistory20 languages (iPhone)REST API · CLI · MCP · private beta
Screens: current build. Files you pick on your iPhone are analysed on the device unless the optional, off-by-default OpenAI provenance check is switched on; public links and web uploads are processed temporarily on our servers.
Web app · web.verifyco.app
On the web, a report a desk can hand over.
- Upload or paste a linkPrivate files up to 250 MB, or a public post from the major social networks.
- Evidence for each of five checksEach check shows what it found, what did not apply, and how much of the file it covered.
- A report you can hand overExport to PDF, share a link, delete on demand. Built for desks that must show their work.
- One account, every deviceSign in with Apple carries iPhone Pro to the web.
05 · How it works
Five evidence families. One honest answer.
- 01Content CredentialsC2PA source claims and signatures
- 02MetadataCamera, software and file context
- 03AI visual evidenceOur in-house visual model
- 04Temporal evidenceChanges across sampled video frames
- 05Frequency evidenceSpatial patterns and compression traces
Evidence contract
Evidence, not a vote
No single family decides alone. When the evidence is thin, the answer stays inconclusive. One written contract says what counts as evidence, what may decide, and what we say when we do not know. The same rules apply on iPhone and on the web.
Report
Verdict band, score and coverage per family
Inconclusive results say which way they lean. “Authentic” needs capture evidence: valid Content Credentials, an in-app capture or an intact camera original. A visual score alone never can.
Opt-in provider checksPDF review, diagnostic onlyUnreadable file: explained, not charged
Capabilities are described at the level of outcomes. Mechanisms are not disclosed.
06 · Who it serves
Consumers today. Professional desks next.
- Today · Consumers and creatorsSelf-serve on iPhone, in 20 languages, and on the web. Free to try.
- Beachhead · EU newsrooms and fact-checkersThey must label AI content under AI Act Article 50 from August 2026. Trust-and-safety and fraud teams follow.
- Expansion · Professional teamsJournalists, trust and safety, fraud and KYC, legal, insurance and forensic labs, through team plans and a metered API.
Our wedge
Verifyco shows its work.
→
Reviewable evidence
Coverage and uncertainty on every check, not one opaque score.
→
One evidence contract
On-device on iPhone, server-side for web, CLI and API: the same rules everywhere.
→
Consumer app and B2B API
The same engine serves a person with one photo and a desk with a queue.
→
Reads the marks that exist
Content Credentials today; supported provider signals with consent. Own-mark verifiers read only their own mark.
07 · What's next
From a live product to an independently validated one.
Establish
- Self-serve API keys and a hosted MCP server
- An independent benchmark on unseen generators, published with confidence intervals
- Android
Validate
- Design-partner pilots with newsrooms, fact-checkers, trust-and-safety and fraud teams
- Metered API and team plans generally available
- Broader Content Credentials signer coverage; work toward C2PA conformance
Scale
- Pilots converted into annual contracts
- Expansion across the EU, the Mediterranean and MENA
- No accuracy claim until a locked, independent corpus with confidence intervals exists
08 · Founder
A solo technical founder who shipped iPhone, web and the engine in six months.
Yusuf Eser, Founder & CEO. Istanbul, Türkiye.
- Before Verifyco
- Surgical technician by training (Fenerbahçe University, 2022): verify before you act. Designing Verifyco's architecture and evidence policy since 2024.
- What was built
- The iPhone app, the web app with billing, the backend, the evidence engine, the visual model, the CLI and the MCP server, written and reviewed by the founder with AI agents as a multiplier.
- Honest by design
- Ships behind automated tests and blocking release gates, and states model limitations openly.
Investors · partners · press
Want the full deck or a live demo?
One email. We reply with the current deck, a demo link and the numbers we share in private.