Content Signals & C2PA: Controlling AI Usage

Content Signals give you a standardized way to declare how automated systems may use your content — more granular than the simple Allow/Disallow of robots.txt. As more and more AI crawlers comb the web, this answers a question that was long open: not only whether but for what a bot may use your site. Content Signals is Cloudflare's implementation of a new Content-Signal directive. This article complements the technical GEO side — the matching frame comes from Structured Data and Technical GEO.

The three Content-Signal categories

The Content-Signal directive works with three categories, each set to yes or no. It cleanly separates classic search, AI training and real-time AI usage:

Signal	Meaning
`ai-train`	Training or fine-tuning AI models on your content
`search`	Building a search index and showing results (links and short excerpts) — not AI-generated summaries
`ai-input`	Feeding content into AI models in real time (e.g. RAG, grounding or AI search answers)

An example in robots.txt:

User-Agent: *
Content-Signal: ai-train=no, search=yes, ai-input=no
Allow: /

Two delivery methods: robots.txt or HTTP header

The same preference can be emitted two ways — via robots.txt or as an HTTP response header with identical key=value grammar. The header is useful where a site-wide robots.txt line is too coarse: for individual URLs, for non-HTML resources such as PDFs or images, and for signals injected at the CDN edge.

Content-Signal: ai-train=yes, search=yes, ai-input=yes

Both methods are equivalent and may be combined. I run exactly that on this domain: Content-Signal: ai-train=yes, search=yes, ai-input=yes sits in the robots.txt and ships as an HTTP header on every response — for a knowledge base explicitly built for AI access, anything else would be dishonest. Because setting a preference is voluntary, the SEO & GEO Analyzer reads Content Signals from both sources and surfaces them as an informational signal, without counting them toward the score.

Four default policies

Four predefined policies cover the most common cases — from fully blocked to fully open:

Disallow All — Most restrictive. No access for any purpose; may cause search engines to exclude your site.
Allow Search Only — Permits search indexing and results, but no AI training and no AI input.
Allow Search & AI Input — Permits search and real-time AI usage (such as AI search answers), but no model training.
Allow All — Permits search, AI input and AI training.

Path-specific rules instead of all-or-nothing

The four policies apply to the entire site — for most sites that is too coarse. The directive can also differentiate per path: the path goes right after Content-Signal:, ahead of the signals. That lets you open up an about page completely, keep the blog on search only and close off internal areas entirely:

# Allow unfettered access to your /about page
User-Agent: *
Content-Signal: /about ai-train=yes, search=yes, ai-input=yes
Allow: /about

# Search-only access to your blog
User-Agent: *
Content-Signal: /blog/ ai-train=no, search=yes, ai-input=no
Allow: /blog/

# Disallow access to your dashboard
User-Agent: *
Disallow: /dashboard/

The same works bot by bot: several User-Agent blocks, each with its own Content-Signal line, grant different crawlers different rights. That is where the practical leverage sits — a blanket line across the whole domain forces you into a compromise you do not actually want anywhere.

The fourth field: `use` — in testing since July 2026

On 1 July 2026 Cloudflare extended the directive with a fourth, optional field: use. It answers a different question than the first three. Those govern what a bot may use your content for; use governs what it may keep and reshare afterwards. Three levels, from restrictive to permissive:

Value	Meaning
`use=immediate`	Interact, but store and reuse nothing
`use=reference`	Index, excerpt, link back — the default value
`use=full`	Summarize and reproduce in full

It is appended to the same line:

User-Agent: *
Content-Signal: search=yes, ai-train=no, use=reference
Allow: /

Two caveats worth knowing. First, this is explicitly a test: since 1 July Cloudflare automatically appends use=reference to every robots.txt it manages, but the generator on contentsignals.org still does not know the field — the policy page continues to document only the three original signals. Second, use remains a preference and not a block, like the other three. Cloudflare does tie it to its bot verification, though: a bot abusing the signals loses its "Verified" status, and bots that reproduce content in full cannot obtain it at all.

Why this matters for GEO

Content Signals are a strategic GEO tool because they decouple visibility from protection. You can allow ai-input — citations in AI search answers — while blocking ai-train: maximum visibility without giving away your content as training data. Four points make the lever:

Strategic visibility — Get cited deliberately without surrendering training data.
Content control — Declare your AI preferences explicitly instead of hoping AI companies respect informal requests.
EU rights reservation — Content Signals include an explicit reservation of rights under Article 4 of EU Directive 2019/790 (opens in a new tab) (Copyright in the Digital Single Market).
Early adoption — As more AI systems honor Content Signals, an early, clear permission record becomes more valuable.

C2PA: provenance, not just permission

Where Content Signals govern how your content may be used, C2PA — the Coalition for Content Provenance and Authenticity — tackles the other direction: it proves where content came from and whether it was altered. Through cryptographically signed metadata (Content Credentials), the provenance of an image or document can be demonstrated. For GEO this is the logical complement to authority: making provenance verifiable provides a trust signal that AI systems are likely to treat as proof of authenticity in future. Both standards are young, and their adoption is still growing.

How to generate your Content Signals

The easiest way to produce the robots.txt block is the Content Signals Generator (opens in a new tab): pick one of the four default policies, customize per category and copy the output straight into your robots.txt. The generator covers the three original signals — the path notation and the new use field you add by hand afterwards. The background is explained in the Cloudflare blog on the Content Signals Policy (opens in a new tab), the extension in the post "Your site, your rules" (opens in a new tab).

FAQ

Does a Content-Signal directive force AI crawlers to obey?

No, it is a declaration, not a technical enforcement. As with robots.txt, the effect depends on a crawler respecting the directive. The value lies in clarity and legal backing: with the EU rights reservation under Article 4 you document an explicit declaration of intent — and create a basis you can invoke.

Does setting signals hurt my AI visibility?

That depends on what you set. Block everything wholesale and you risk vanishing from search and AI answers. The GEO-friendly variant is differentiated: allow search=yes and ai-input=yes — which keeps you citable — and block only ai-train as needed. That way you steer usage without sacrificing visibility.

What changes at Cloudflare on 15 September 2026?

Preference turns into enforcement there — and you should keep those two layers strictly apart. On 15 September 2026 Cloudflare switches its defaults: on pages carrying ads, crawlers in the Training and Agent categories will be blocked by default, while Search stays allowed. This applies to newly onboarded domains and to existing free customers who have not touched their settings by then. A Content-Signal line is a request; this is a block at the network level that a crawler cannot simply ignore.

The catch is in the evaluation logic: the most restrictive applicable rule wins. Multi-purpose crawlers such as Googlebot, Applebot and BingBot crawl for search and for training — block training and you shut them out entirely, classic search indexing included. If your site sits behind Cloudflare, look into your security settings before 15 September rather than being surprised by a drop in visibility. The full switch — taxonomy, deadlines and a checklist — is covered in 15 September 2026: Cloudflare Changes the Crawler Defaults.

Do I need C2PA if I only publish text?

Rarely essential today, sensible in perspective. C2PA shows its strength above all with images, video and audio, where authenticity and manipulation become an issue. For plain text, verifiable authority — author pages, schema, transparent sourcing — weighs more heavily today. Still, watch the development, because provenance proofs will gain importance as a trust signal.