AI System Model Card

SemlyPro Canonical location: https://semlypro.com/ai-transparency/model-card

Important notice — informational only; no warranty; operator identity

This model card is provided for transparency purposes only. It describes the intended design, behaviour and controls of the SemlyPro AI System. It is not a warranty, guarantee, performance representation or contractual commitment, and it does not create or vary any legal obligation. It describes design intent, not guaranteed outcomes. Model and system behaviour can change, and the descriptions here may become out of date between updates.

The governing commitments, disclaimers and limitations of liability that apply to your use of the Service are set out exclusively in the Terms of Service (https://semlypro.com/terms), the Acceptable Use Policy (https://semlypro.com/acceptable-use), the Data Processing Agreement (https://semlypro.com/dpa) and, for enterprise customers, the Master Services Agreement. If anything in this card appears to conflict with those documents, those documents prevail.

Operator. The Service is currently operated by Semly Pro, a sole proprietorship (eenmanszaak) established in the Netherlands, registered with the KvK (Dutch Chamber of Commerce) under number 99448351, VAT ID NL005387029B31, registered address Hawaiiweg 41, 1339 NW Almere, Netherlands (trading as “SemlyPro”, “we”, “us”, “our”). SemlyPro intends to incorporate a private limited company (Semly Pro B.V.); on incorporation, Semly Pro B.V. will assume the operation of the Service and references to “SemlyPro” will be read as references to Semly Pro B.V. Nothing in this card is legal advice.

Table of Contents

1. What this card describes
2. System overview
3. AI providers, external AI engines and connected data integrations
4. Data used at inference time
5. Data used to train the underlying models
6. Evaluations and quality processes
7. Known limitations
8. Safety controls
9. Human oversight
10. EU AI Act classification and role
11. Compliance framework
12. Feedback and reporting
13. Version history

1. What this card describes

This model card describes the SemlyPro AI System — the collection of features and pipelines through which SemlyPro generates, analyses, and publishes content on behalf of customers.

SemlyPro does not train its own general-purpose AI models. The SemlyPro AI System is built by composing third-party general-purpose models with SemlyPro’s own retrieval pipelines, prompt engineering, evaluations, and content-marking layer.

2. System overview

Purpose. Help businesses rank on Google and be cited in AI search engines by generating SEO-optimised content that is grounded in real competitor research and by tracking citation performance across search engines and AI systems.

Intended users. SEO professionals, marketing teams, content teams, agencies, and product-led-growth teams at B2B and B2C companies operating in EU, UK, India and global markets. Multi-tenant SaaS with per-account isolation.

Intended use cases.

• Keyword research and content-brief generation.
• Long-form content generation with human review.
• Multi-CMS one-click publishing.
• SEO audit and site optimisation recommendations.
• AI-search citation tracking (across ChatGPT, Perplexity, Google AI Overview, Gemini and Claude).

Out-of-scope use cases (prohibited). See the SemlyPro Acceptable Use Policy at https://semlypro.com/acceptable-use, in particular:

• section 3.7 (Article 5 EU AI Act prohibited practices);
• section 3.8 (Annex III EU AI Act high-risk uses);
• section 3.9 (political and electoral content);
• section 3.10 (impersonation and rights of publicity); and
• section 3.6 (regulated professional / YMYL sectors).

(These section numbers follow the renumbered Acceptable Use Policy, in which the EU-AI-Act prohibited-use blocks are numbered 3.7–3.10 and “regulated professional sectors” is section 3.6.)

3. AI providers, external AI engines and connected data integrations

3.1 Generation providers

The SemlyPro AI System integrates general-purpose AI models from the following providers for content generation and analysis. Model versions are pinned in production and updated on a controlled schedule.

Model version changes are recorded in the SemlyPro changelog. Significant behavioural changes are notified to customers by email. The pinned-version table above is maintained under change control and updated whenever a version is re-pinned.

Model Context Protocol (MCP). SemlyPro operates a write-capable MCP server usable from compatible clients (for example, Claude Desktop / Claude Code), through which an external agent can trigger content generation and, where enabled, CMS publishing. MCP write actions that have external effects (in particular publishing) require explicit prepare-then-confirm acknowledgement, remain subject to the AUP rate limits and the human-in-the-loop expectations described in sections 8 and 9, and cannot perform fully autonomous significant actions without the customer’s explicit enablement.

3.2 External AI engines queried for citation / brand tracking

For the AI-search citation and brand-tracking feature, SemlyPro sends queries (which may include brand names, target keywords and customer-supplied inputs) to external AI answer engines to measure whether and how a brand is cited. These engines are not generation providers, but they receive query data and are therefore third-party recipients. They currently include ChatGPT (OpenAI), Perplexity, Google AI Overview / Gemini (Google), and Claude (Anthropic). The data sent to each engine is limited to what is necessary to run the tracking query. See the Sub-processor List at https://semlypro.com/subprocessors for the authoritative, current list of recipients.

3.3 Connected customer data integrations

Where a customer connects them, the SemlyPro AI System ingests data from the following customer-authorised integrations (Google Ireland Limited / Google LLC), which the customer authorises by OAuth:

• Google Ads — campaign, spend, performance and conversion data.
• Google Analytics 4 (GA4) — traffic, audience and event data.
• Google Search Console (GSC) — query, impression, click and index data.

SemlyPro also processes customer-uploaded Excel / CSV files — arbitrary tabular data the customer imports, which may contain personal data (in which case the customer is the controller / Data Fiduciary). SemlyPro’s use of Google user data obtained through these integrations is governed by Google’s API Services User Data Policy, including its Limited Use requirements; see the Google API Services / Limited Use Compliance Statement and the Privacy Policy at https://semlypro.com/privacy. These four sources are treated as Customer Materials where customer-controlled (see section 4).

4. Data used at inference time

At inference (content generation and analysis), the SemlyPro AI System uses:

• Customer Materials provided by the requesting customer (target keyword, brand voice, brief inputs, connected competitor URLs).
• Connected integration data, where the customer has connected it and where it is used for the requested analysis — namely Google Ads, Google Analytics 4 and Google Search Console data (section 3.3), and data from customer-uploaded Excel / CSV files. Where such data is used to ground SEO-audit and recommendation features, it may be transmitted to the generation providers named in section 3.1 as part of the analysis; where a feature does not send this data to the models, that data is used only within SemlyPro’s own pipelines.
• Grounding data scraped by SemlyPro from public web pages ranking for the target keyword.
• Cached prior outputs in your own account, where you have referenced them (for example, when regenerating a piece).

At inference, we do not use:

• Customer Materials or Content of other customers.
• Personal data from third parties beyond what is included in publicly available scraped web content.

Transmission to the AI providers. Customer Materials and other inference inputs described above are transmitted to OpenAI Ireland Limited and Anthropic PBC under commercial / enterprise API terms that (a) prohibit the use of inputs or outputs to train those providers’ publicly-released models and (b) apply zero or limited data retention. See the Sub-processor List (https://semlypro.com/subprocessors) and the Data Processing Agreement (https://semlypro.com/dpa).

5. Data used to train the underlying models

SemlyPro does not train the underlying general-purpose AI models. The AI providers train those models using their own training data. Under EU AI Act Article 53(1)(d), providers of general-purpose AI models must publish a sufficiently detailed summary of the content used to train the model, according to a template published by the AI Office. (Article 53(1)© is the separate obligation to put in place a policy to comply with Union copyright law.) The AI providers’ training-content summaries are published at:

• OpenAI — see openai.com/policies and related transparency reports.
• Anthropic — see anthropic.com/legal and related transparency reports.

We do not use your data to train publicly-released AI models. Customer data and outputs may be used internally and non-publicly to evaluate, debug, secure and improve the SemlyPro AI System (see sections 6 and 9), which is distinct from training public models. Full detail is in our AI Transparency Notice section 11 and Privacy Policy section 5; the wording here is intended to match those documents.

5.1 Personnel access and internal use of Customer Materials

SemlyPro personnel and authorised contractors may access Customer Materials and Content only as necessary to: (a) provide, maintain and secure the Service; (b) troubleshoot, debug and provide support; © investigate abuse, security or AUP issues; and (d) review and improve the Service and its models for SemlyPro’s internal, non-public purposes only. Such access is subject to confidentiality obligations, least-privilege / role-based access controls, and access logging. SemlyPro does not use Customer Materials or Content to train publicly-released AI models. This clause mirrors the equivalent provisions in the Terms of Service, Privacy Policy and DPA.

6. Evaluations and quality processes

The SemlyPro AI System is evaluated across the following dimensions. The figures below are indicative, informational, and not warranties (see the “Important notice” above); each must be independently defensible and dated before it is published.

Where indicative figures are not published above, the detailed current benchmark numbers are made available under NDA to enterprise customers on request to anil@semlypro.com. Any figure published here is provided for transparency only and creates no warranty or contractual representation.

7. Known limitations

• Hallucination. The AI System may fabricate facts, statistics, quotes, prices, product features, and attributions. This is inherent to current large-language-model technology. You must review outputs before publication.
• Bias. Generated content may reflect biases present in the underlying training data of the AI providers, in the scraped grounding sources, or in Customer Materials.
• Currency. The AI providers’ models have training-data cut-off dates. Content about very recent events may be inaccurate or absent.
• Language coverage. Quality varies by language. We provide fluent output in EU languages (Dutch, English, German, French, Spanish, Italian, Portuguese, Polish and similar), with degraded output quality for less-common languages.
• Domain-specific accuracy. Highly technical domains (medical, legal, financial, scientific) require human expert review.
• Character limits and structural output. Very long-form generation (>5,000 words in one pass) may lose coherence.

8. Safety controls

The SemlyPro AI System applies the following safety controls:

• Input filters that refuse prompts violating the AUP or triggering prompt-injection signatures.
• AI-provider safety filters applied by OpenAI and Anthropic at the model layer.
• Output filters that refuse to return content in the categories prohibited by our AUP.
• Rate limits and per-domain caps to reduce abuse and reduce risk of search-engine penalisation for customers.
• Confirmation-gating of write actions, including MCP-initiated publishing, which requires explicit prepare-then-confirm acknowledgement (see section 3.1) and remains subject to the AUP.
• Human-in-the-loop by design — all long-form Content is intended for human review before publication.
• Content marking via machine-readable markers in output, to be live by the applicable EU AI Act deadline. The Article 50(2) machine-readable marking obligation applies from 2 August 2026; generative-AI systems already placed on the market before that date benefit from a transitional period to 2 December 2026 to meet the Article 50(2) marking requirement under the May 2026 AI Omnibus agreement (agreed 7 May 2026; text 13 May 2026), which remains subject to formal adoption. SemlyPro will implement marking by the earlier of its own product timeline and the applicable statutory deadline.

Controls are updated when new attack patterns are observed. The descriptions above state design intent, not guaranteed outcomes (see the “Important notice”).

9. Human oversight

The SemlyPro AI System is designed for human-in-the-loop use. Every long-form Content item is intended for review, edit, and approval by a human before publication. Automated publishing — including publishing initiated via the MCP server — is available only for shorter Content types, only where the customer has explicitly enabled it, and only subject to the confirmation-gating, rate limits and safeguards in the AUP (see sections 3.1 and 8).

We do not provide, and do not support, use of the AI System for fully autonomous decision-making with legal or similarly significant effects on natural persons (see Privacy Policy section 5.3, GDPR Article 22 and UK GDPR Article 22 / Data Protection Act 2018). For Indian Data Principals, the System is likewise not intended to make automated significant decisions about individuals; SemlyPro’s processing is subject to the notice-and-consent and Data Fiduciary obligations of India’s Digital Personal Data Protection Act 2023 (see section 11).

10. EU AI Act classification and role

For the purposes of the EU AI Act (Regulation (EU) 2024/1689), SemlyPro’s role is twofold:

• SemlyPro is a deployer of the AI systems it builds on the general-purpose AI models supplied by OpenAI and Anthropic; and
• SemlyPro is also a provider of its own downstream AI system that generates synthetic text content. As such, SemlyPro bears the Article 50(2) transparency / machine-readable-marking obligations directly (see section 8), within the Article 25 value-chain framework.

The SemlyPro AI System is not intended for use in any Annex III high-risk category. Use for those categories requires SemlyPro’s prior written consent and is prohibited by the AUP (see section 2).

11. Compliance framework

The SemlyPro AI System’s design and operation is informed by the following instruments across its EU, UK and India markets. This list is descriptive of the framework SemlyPro takes into account; it is not an admission that every instrument applies to every activity.

European Union

• EU AI Act (Regulation (EU) 2024/1689) — Articles 5, 25, 26, 27 (where applicable), 50, 53(1)(d) and 73.
• GDPR (Regulation (EU) 2016/679) — Articles 22, 32.
• Digital Services Act (Regulation (EU) 2022/2065) — Articles 16 and 20 where, and only to the extent that, SemlyPro operates a user-facing hosting or online-platform service. SemlyPro’s core B2B SEO generation tool is unlikely to be a “hosting service” or “online platform” for the content it helps customers produce; this citation does not imply a voluntary assumption of platform-level DSA duties.
• Regulation on the transparency and targeting of political advertising (Regulation (EU) 2024/900) — where users generate political content.
• Voluntary EU codes of practice relevant to Article 50 transparency and to general-purpose AI (as issued by the EU AI Office), to the extent applicable.

United Kingdom

• UK GDPR and the Data Protection Act 2018 — including Article 22 UK GDPR for automated decision-making (see section 9).
• Privacy and Electronic Communications Regulations 2003 (PECR) — where electronic marketing or cookie-based tracking is involved (see the Cookie Policy).
• Online Safety Act 2023 — SemlyPro’s assessment is that its core B2B product is out of scope because generated Content is published to the customer’s own external CMS rather than shared user-to-user among SemlyPro users; any public user-to-user or shared-output surface (for example, free tools on the marketing site) is separately scoped.

India

• Digital Personal Data Protection Act 2023 (DPDP Act) and the Digital Personal Data Protection Rules 2025 (notified 13–14 November 2025; the Data Protection Board of India is being instituted, with core Data Fiduciary obligations operative on and from 13 May 2027). SemlyPro acts as a Data Fiduciary (and as a Data Processor where the customer is the Data Fiduciary), and provides notice-and-consent, Data Principal rights, and a Grievance Officer (anil@semlypro.com). The DPDP Act does not contain a GDPR-style automated-decision article, but it imposes notice, consent and Data Fiduciary obligations relevant to AI processing, and heightened protections for children (persons under 18), including verifiable parental consent and no tracking or targeted advertising directed at children. See the India Privacy & Grievance Addendum and the Privacy Policy.

12. Feedback and reporting

Use the routing below (a single package-wide routing table applies across all SemlyPro legal documents):

• Model behaviour issues, hallucination reports and AI/model-quality concerns: anil@semlypro.com
• Serious safety or security incidents (including AI incidents): anil@semlypro.com
• AI transparency questions and formal legal notices: anil@semlypro.com
• India grievances / Data Principal complaints (DPDP Act, IT Rules 2021, Consumer Protection (e-Commerce) Rules 2020): anil@semlypro.com

13. Version history

This version-history row is kept current on every material change to the AI System or to this card.

Last updated 13 July 2026 · Operated by Semly Pro (eenmanszaak), KvK 99448351, Hawaiiweg 41, 1339 NW Almere, Netherlands · Questions about this document: anil@semlypro.com