Generated reference: edit sources under docs/helix/, then npm run site:build.

Generated from HELIX docs. Source: docs/helix/00-discover/competitive-analysis.md. Edit the source, not this file or site/ HTML.

Source frontmatter

ddx:
  id: aibadge.competitive-analysis
  type: competitive-analysis
  links:
    - target: aibadge.research.prior-art
      kind: informed_by
    - target: aibadge.product-vision
      kind: informs
    - target: aibadge.prd
      kind: informs
status: draft
activity: 00-discover
created: 2026-07-22

Competitive Analysis: Content Origin Labeling and Provenance

Scope: Technologies and products that label, watermark, or cryptographically attest how digital content was made, or who stands behind a claim about it. Operator use cases under test:

UC-AI-docs: Long-lived documentation sites (e.g. HELIX microsite): mostly

model-generated; maker wants a signed colophon that proudly names tools (e.g. Claude) and binds the claim to a house or person.

UC-human-social: Short posts on X/Twitter and similar: maker wants it

clear the words are human-authored, under their control, not a platform "Made with AI" stamp or a purity cult badge alone.

Market Landscape

Attribute	Assessment
Market Maturity	Emerging / fragmenting: industrial standards, vendor watermarks, and DIY badges coexist; no single consumer mental model
Key Trends	Layered AI-image labeling (C2PA + SynthID); EU AI Act Art. 50 transparency (from 2 Aug 2026); human-only badge movements; platform-side labels
Entry Barriers	Low for badges and claim formats; high for becoming the media-industry default (C2PA coalition + camera OEMs)
Buyer Power	High for individual makers (many free substitutes); low influence over platform strip of metadata

Layer map (what competes for the same job)

Layer	Job	Examples	Fits UC-AI-docs?	Fits UC-human-social?
A. Craft / social badge	Visible origin story the maker chooses	Not By AI, ad-hoc footers	Partial (no proud-AI peer)	Partial (human-only, no crypto)
B. Signed media provenance	Tamper-evident file history	C2PA Content Credentials, CAI SDK	Partial (docs/images if file-bound; heavy)	Weak (uploads strip metadata; text posts lack files)
C. Model-side watermark	Survive distribution without rich payload	SynthID, Meta Video Seal	Weak (vendor, not maker BOM)	Weak (no human-authored path)
D. Metadata taxonomy	Structured source codes in EXIF/IPTC/XMP	IPTC `digitalSourceType`	Partial (vocabulary only)	Weak (stripped on social)
E. Platform disclosure	Site labels AI media	Meta, others	N/A	Platform-owned, not portable
F. Detectors	Guess AI vs human	Various classifiers	No (wrong job)	No (wrong job)
G. Prose quality / AI-tell edit	Detect or rewrite AI-shaped writing patterns	Sloptimizer (`easel-skills`)	Complementary (edit step, not origin seal)	Complementary (does not declare human authorship)

Thesis implication: UC-AI-docs needs A+B-like maker signing with a rich BOM (not only vendor watermark). UC-human-social needs A that survives metadata death (visible mark + outlink), with human composition first-class. No incumbent owns both as one brand system.

Adjacent: Sloptimizer (complement, not rival)

Source: /home/erik/Projects/easel-skills/skills/sloptimizer (Easel skill; Vale-backed detect / rewrite / validate). Shared house context with HELIX/DDx sample.

Dimension	Sloptimizer	Innsigle
Job	Remove AI-isms, filler, unsupported claims; harden specs	Declare colophon; optional sign who sealed it
Authorship score	Explicitly does not claim text was written by AI	Explicitly is not a detector
Output	Cleaner prose / executable work items	Mark + bill + optional signature
UC-AI-docs fit	Post-generate edit pass before publish	Seal names models (e.g. Claude) and can list editorial tools
UC-human-social fit	Optional polish if human draft has AI-tell patterns	Human-authored mark only when composition is actually human

Pipeline (sample target for methodology docs):

model draft (Claude, …) → sloptimizer detect|rewrite|validate → Innsigle colophon
  (model-primary + named tools + human roles + optional "sloptimizer" tool line)
  → house sign → publish site

Hard boundary (product integrity): A sloptimizer rewrite does not turn model-primary text into human-authored for Innsigle. Sounding less AI is a prose outcome; composition state tracks how the words were produced, not how human they read after edit. Mislabeling AI→sloptimizer text as human is a category violation (same class as detector theater).

Why this matters competitively: Not By AI and detectors both push toward a binary human/AI appearance test. Operators can run both in sequence: use models, clean the prose, disclose the bill, sign it. That is a different category story than "make it look human so no one notices."

Competitive Forces

Force	Pressure	Evidence	Implication
Direct rivalry (badges)	Med	Not By AI has recognition and media coverage	Do not compete as human-only sticker clone
Direct rivalry (standards)	High in media	C2PA / CAI backed by Adobe, camera OEMs, major AI labs	Do not position as C2PA replacement; optional bridge later
Substitutes	High	Footnotes, README lines, platform labels, silence	Product must be faster/clearer than a sentence in a footer or add verify depth footnotes lack
New entrants	High	Low cost to ship SVG + docs	Defensibility = mark recognition + claim format + sample on real properties
Buyer power	High	Makers switch costs near zero	Sample (Azgaard, HELIX docs, social) is the adoption path

Competitor Profiles

Competitor	Type	Positioning	Strengths	Weaknesses vs Innsigle UCs	Confidence
Not By AI	Direct (brand badge)	Human-made pride badge; 90% rule; project pages	Clear mark; low friction; cultural recognition	Human-only; self-asserted; no crypto; no proud-AI BOM for UC-AI-docs	High (product site)
C2PA / Content Credentials (CAI)	Indirect (standard + tools)	Nutrition label for media origin and edits; signed manifests	Open standard; industry adoption; CAI open-source SDK; verify UIs	File/media-centric; fragile when metadata stripped; identity optional; not a dual craft brand; weak for plain text social	High
SynthID (Google; also OpenAI images)	Substitute (vendor watermark)	Invisible model-origin signal in media	Survives some transforms; pairs with C2PA in vendor stacks	Not maker-authored BOM; little payload; no human path; not text-post native	High
Meta Video Seal	Substitute	Open-source video watermark research/product line	Open direction for video durability	Video-focused; not dual human/AI craft seal for docs+social	Med
IPTC digitalSourceType	Indirect (taxonomy)	Codes for trained algorithmic media, composites, etc.	Shared vocabulary for news/media metadata	Not a brand; not a signer UX; rides on metadata that social strips	High (IPTC docs)
Platform labels	Substitute	"Made with AI" style UI	Audience already sees them	Not maker-controlled; not portable; no personal key	High
AI detectors	False substitute	Score content as AI/human	Buyer demand for certainty	Arms race; wrong product category; liability	High
DIY footer / README disclosure	Substitute	Free text declaration	Zero cost	No mark system; no verify; easy to ignore	High
Sloptimizer (easel-skills)	Complement	Audit/rewrite AI-shaped prose; does not claim authorship	Vale rules; HELIX/DDx adapters; house-aligned	Not an origin seal; no social mark; no DKIM-like bind	High (skill contract + repo)
Bare Sigli / consultancies	Name collision only	AI software shops	SEO noise	Not provenance products	High (sigli.com)

Feature Comparison

Capability	Innsigle (target)	Not By AI	C2PA/CAI	SynthID	IPTC DST	Platform label
Visible maker mark	Full	Full	Partial (pin UI)	None	None	Platform UI
Human-authored state	Full	Full	Partial (if asserted)	None	Partial codes	Rare
Proud AI / model-named BOM	Full	None	Partial (tool in manifest)	Weak signal	Partial codes	Weak
Same system for human + AI	Full	None	Yes (neutral)	No	Yes (codes)	No
Maker-controlled crypto sign	Full (DKIM-like)	None	Full (cert chain)	N/A	None	None
Survives social metadata strip	Partial (visible mark + URL)	Partial (image badge)	Weak	Strong for media	Weak	N/A
Text / HTML docs site	Full	Partial	Partial	Weak	Weak	None
Text post on X/Twitter	Partial (badge/link pattern)	Partial	None native	None	None	Platform only
Not a detector	Full	Full	Full	Detection-side	Full	Full

Legend: Full / Partial / None / Weak relative to the two operator UCs.

Durability and channel reality

Channel	What survives	What dies	UC impact
Owned docs site (Hugo/static)	HTML, footer mark, linked claim, signed sidecar or well-known URL	Little	UC-AI-docs: full signet + rich BOM is viable
Downloaded PDF/image with C2PA	Manifest until re-export	Often lost on screenshot/re-encode	Optional later interop, not P0
X/Twitter text post	Visible characters, attached image if any	Embedded file metadata routinely	UC-human-social: must be visual + short link or image card; no file-manifest dependency
X/Twitter image post	Pixels (watermark may live); badge in image	C2PA often stripped	Prefer mark baked into image or clear caption + verify URL

Industry practice (2025–2026): vendors stack C2PA (rich, fragile) with SynthID (sparse, durable) for images. That stack does not give a maker a named Claude BOM on a docs site or a human seal on a tweet.

Differentiation Strategy

Differentiator	Why it matters for UCs	Defensibility
Dual composition brand (human and AI equally first-class)	UC-human-social and UC-AI-docs share one mark family	Med (brand + guidelines)
Colophon with named tools (e.g. Claude) as pride, not confession	UC-AI-docs needs model credit without shame language	Med (vocabulary + UX)
Maker/house signet (DKIM-like), not only vendor watermark	UC-AI-docs: "Azgaard sealed this bill"	Med-High if key discovery is simple
Social-safe pattern (mark + outlink; no metadata dependence)	UC-human-social on X	Med (design pattern, not a protocol monopoly)
Explicit non-detector, non-C2PA-replacement stance	Avoids arms race and standards war	High (positioning discipline)

Positioning (sharpened): For operators who publish both model-heavy docs and human social posts, Innsigle is a content-origin seal and colophon: one visual system for human-authored and AI-generated work, with optional personal/house signatures. Unlike Not By AI, AI is a first-class bill ingredient. Unlike C2PA and SynthID, the primary job is maker-facing brand and claim for web text and social, not industry media pipelines or model-side watermarks.

Strategic Implications

Move	Action
Attack	Owned documentation sites with signed AI colophons (sample HELIX / Azgaard docs); optional sloptimizer pass recorded in the bill; human social mark packs for X-class platforms; CLI sign/verify for site content
Defend	Category integrity (not detection; not "sloptimized = human"); dual-state mark family; simple key custody for house signers
Avoid	Competing as the C2PA implementation of record; building detectors; human-only purity as the only story; requiring C2PA survival on Twitter for P0; treating prose de-AI-ifying as composition change to human-authored
Interop (later)	Map colophon fields to IPTC `digitalSourceType` / C2PA assertions when exporting media files; keep Innsigle claim as the maker layer; optional skill/CLI hook after sloptimizer in easel-skills or agent workflows

Evidence anchors (survey)

C2PA: open standard; Content Credentials "nutrition label"; CAI open-source tools

(contentauthenticity.org, c2pa.org).

SynthID vs C2PA: metadata-rich vs pixel-durable; OpenAI documents both on images

(2026 public materials).

Not By AI: human-centric badge; not a detector (notbyai.fyi).
IPTC NewsCodes digitalSourceType: algorithmic / composite codes for media

metadata (iptc.org).

EU AI Act Art. 50: transparency duties for synthetic content (from 2 Aug 2026).
Channel fact: social re-encoding commonly strips signed metadata (industry

commentary and CAI durability discussions); treat as design constraint for UC-human-social.

Open research (non-blocking)

Exact claim envelope vs C2PA assertion subset (design).
Whether a tweet-length "claim pointer" URL scheme is enough for social verify.
Trademark/name collisions beyond Sigli (ongoing).