AI Visibility Platform Security Checklist: ISO 27001, SOC 2, GDPR and DPAs Explained (2026)

Key takeaways

• ISO 27001 is a certificate, SOC 2 is an auditor's report, and GDPR is a law. A vendor can hold the first two and still have a weak DPA, so check all three.
• ISO 27001:2013 certificates expired on October 31, 2025. Ask for a 2022-edition certificate, its scope, and the Statement of Applicability.
• AI visibility platforms often ingest server logs, CDN logs, Search Console data and CMS credentials. Ask what they store, for how long, and who can read it.
• A usable DPA names sub-processors (including any model providers), sets breach-notification timelines, and gives you real audit rights.
• Most of this can be verified in one or two email exchanges. Do it before the trial, not after the contract.

Why security matters more for AI visibility tools than you'd expect

A rank tracker used to be low risk. You gave it a domain and a keyword list. Nothing sensitive moved.

AI visibility and GEO platforms are different. The useful ones connect to more of your stack: Google Search Console, Bing Webmaster Tools, analytics events, CDN or server logs, and sometimes a CMS with publish rights. Some also store brand documents, tone-of-voice guides and product catalogs. That is a wider data footprint than most marketing tools carry.

Take crawler analytics. To show which AI bots hit your pages, a platform has to read your Cloudflare, Vercel, Fastly or raw server logs. Logs can contain IP addresses, URLs with query strings, and user agents. Under GDPR, IP addresses can be personal data. So a vendor that sounds like a harmless dashboard is, for compliance purposes, a processor of your visitors' data.

That is the reason to run a proper checklist instead of glancing at a trust badge on the footer.

ISO 27001, SOC 2 and GDPR: what each one actually tells you

The three get lumped together in questionnaires. They answer different questions.

One useful way to think about it: ISO 27001 and SOC 2 are evidence a vendor can show you. GDPR is the obligation you carry as the buyer. You need the first two to help satisfy the third.

ISO 27001: check the edition, scope and SoA

ISO/IEC 27001:2022 was published on October 25, 2022. The transition deadline was October 31, 2025, so every ISO 27001:2013 certificate has now lapsed. A vendor that waves a 2013 certificate at you is not currently certified.

Ask for three things:

1. The current certificate, which should reference the 2022 edition.
2. The certificate scope. It should cover the product and hosting environment you'll use, not just a corporate office or one subsidiary.
3. The Statement of Applicability (SoA), which lists which Annex A controls apply and which were excluded, and why.

The scope check catches more problems than anything else. A certificate covering "HR and office operations" says nothing about the platform processing your logs.

SOC 2: Type II beats Type I

SOC 2 evaluates controls against five Trust Services Criteria: security, availability, processing integrity, confidentiality and privacy. Type I is a snapshot at one date. Type II tests whether controls worked over a period, usually six to twelve months.

Most enterprise procurement teams now ask for Type II. One underwriting source claims 78% of B2B enterprise buyers require SOC 2 before signing, and that figure comes from a compliance vendor, so treat it as directional. Another vendor source puts a first Type II at 9 to 10 months minimum and $50,000 to $200,000 or more, depending on scope. Again indicative, but it explains why young startups often have Type I or "in progress" status.

When you get the report, read the auditor's opinion (qualified or unqualified), the list of exceptions, and the complementary user entity controls, which describe what you are responsible for.

GDPR: the part that sits with you

GDPR does not certify anyone. If you're the controller and the platform processes personal data for you, you need a DPA under Article 28, and you need confidence that the vendor applies appropriate technical and organizational measures under Article 32.

The frameworks overlap heavily, which is good news for vendors and for you. DSALTA's control mapping shows access control landing in SOC 2 CC6.1 to CC6.3, ISO 27001 A.5.15, A.5.16 and A.5.18, and GDPR Article 32. Logging and monitoring map to SOC 2 CC7.2 and CC7.3, ISO 27001 A.8.15 and A.8.16, and GDPR Articles 30 and 32. Another source puts the SOC 2 and ISO 27001 control overlap at 60 to 70%. So a vendor with one certification has done much of the work for the other, which is a fair reason to be suspicious when they have neither.

The DPA: eight clauses worth reading

Teams often file the DPA as legal paperwork. Vanta's guide to GDPR-compliant DPAs makes the right point: treat it as an operational document. A DPA is a binding contract between controller and processor, and Article 28 spells out what it must cover.

Here is what to check in an AI visibility vendor's DPA, in plain terms.

Processing instructions and scope

The DPA should say what data is processed, why, and for how long. For a GEO platform that might be: prompt and response data, site analytics events, crawler logs, Search Console data, and user account details. If the vendor lists "all data you provide" with no categories, push back.

Confidentiality

Everyone with access to your data, including employees, contractors and temporary staff, must be bound by confidentiality. Good DPAs say the obligation survives after someone leaves. Ask about role-based access and onboarding and offboarding procedures too.

Security of processing

Article 28 points to Article 32. The DPA should reference actual measures, not "industry standard security". Vanta's list of common drafting mistakes is worth remembering: generic security language, no named standards, and no provision for ongoing oversight. Look for encryption in transit and at rest, access controls, logging, MFA, and a stated incident process.

Sub-processor management

This is where AI tools differ most from older software. Vanta splits AI-relevant sub-processors into three groups:

1. Infrastructure sub-processors, such as the cloud host.
2. Model providers, when the vendor wraps OpenAI, Anthropic or another foundation model through an API.
3. Data enrichment or annotation vendors, where your data is used for fine-tuning or human review.

Ask for the public sub-processor list, a notification period for changes, and your right to object. Then ask the pointed question: is my data used to train any model, theirs or a sub-processor's?

Data subject rights, breach notification and deletion

The vendor must help you answer access and deletion requests. Breach notification should fit inside your own 72-hour window to the regulator, so a DPA that promises notice "without undue delay" with no number gives you little. Also confirm what happens at contract end: deletion or return, within what timeframe, and whether backups are covered.

Audit rights

Processors must allow and contribute to audits. Weak DPAs leave this vague or make it so restrictive it is useless. A fair version accepts independent audit reports (SOC 2 Type II, ISO 27001) as the first line of assurance, defines how evidence is shared, and keeps on-site audits for cause.

International transfers

If the vendor or its sub-processors sit outside the EU, the DPA needs a transfer mechanism, such as standard contractual clauses or participation in the EU-US Data Privacy Framework. I didn't find fresh sourcing on current Data Privacy Framework status, so confirm it directly with the vendor and your counsel rather than relying on a badge.

The practical checklist

Copy this into your procurement doc. Most items can be answered by the vendor's trust center or a single security questionnaire.

Questions specific to AI visibility platforms

A few of these rarely appear in generic SaaS questionnaires.

• If the platform publishes content to your CMS, what permissions does it need, and can you restrict it to drafts?
• Does it keep a log of every action an AI agent takes in your account?
• If you connect Search Console or Bing Webmaster data, can you revoke access without contacting support?
• Are prompts and responses shared across customers, or isolated per workspace?
• Does the vendor publish aggregated research from customer data, and is that anonymised?

The last one matters more than it sounds. Promptwatch Data publishes aggregated, non-identifiable trends from 26B+ analysed data points collected from real AI interfaces. That is a reasonable model: aggregate trends are fine, and you should still ask any vendor to confirm in writing that your own site data is not exposed.

Crawler logs deserve their own review

If you plan to feed AI crawler data into a platform, think about what you're sending. Promptwatch Data's AI crawler traffic report covers June 8 to September 6, 2026 and counts only IP-verified requests from OpenAI, Anthropic, Google, Perplexity and Mistral, so spoofed user agents are excluded. In the week of August 31 to September 6, OpenAI made up 79.8% of verified requests, down from 94.8% in the week of June 8 to 14. Useful context for your own logs.

The security angle is the integration itself. Platforms such as Promptwatch work with Cloudflare, Vercel, Fastly and raw server logs, which means the vendor holds some of your traffic data. Before connecting, ask how much of each log line is retained, whether IPs are hashed or truncated, and how long logs live.

It's also worth checking robots.txt, CDN and WAF rules for each AI provider so you aren't blocking crawlers by accident or exposing paths you meant to keep private. A security review and a visibility review overlap more than people expect.

Comparing vendors: what I'd actually do

The tools catalog for this category is large. I couldn't find a public, side-by-side comparison of security postures across AI visibility platforms, so I won't pretend one exists. Anything you read claiming a ranking of vendors by compliance should be verified against the vendor's own trust page.

Here is the process I'd use for a shortlist of three to five tools, such as Profound, Scrunch, Peec AI, Otterly.AI and Promptwatch:

1. Send the same security questionnaire to each vendor on the same day.
2. Request the ISO 27001 certificate, SOC 2 Type II report and DPA before any call.
3. Rank answers by specificity, not by badge count. A vendor with a short but precise answer on sub-processors beats one with five logos and no list.
4. Have your privacy lead read the DPA, not just procurement.
5. Start with the least sensitive integration (prompt tracking only), and add log ingestion or CMS publishing after the review passes.

Step five is the cheapest risk control in the list. Tracking prompts about your brand moves very little sensitive data. Crawler logs and CMS publish rights are where the real exposure sits, so earn those connections.

For wider browsing, the GEO software directory at bestgeosoftware.com lists tools in this category.

Do you need AI-specific certifications too?

Possibly, depending on how the vendor uses AI. The Aug 2026 DSALTA articles cover ISO 42001, NIST AI 600-1, AIUC-1 and EU AI Act deadlines. I couldn't verify their requirements independently for this guide, so read them before you add these to your questionnaire.

My take: for a tool that monitors AI answers and doesn't make decisions about people, ISO 42001 is a nice signal, not a blocker. If the platform runs autonomous agents that write and publish content on your behalf, ask harder questions about human review, action logs and rollback. Those are governance controls, and they matter more than another logo.

If you're on a tight timeline and need to run this with outside help, an agency such as 1001 SEO Media can run the vendor selection alongside the GEO strategy, but the checklist above works fine in-house.

Putting it together

The lowest-effort version of this review is a one-page scorecard: ISO 27001 (2022, scoped correctly), SOC 2 Type II, DPA with sub-processors named, a training-data statement, and log retention periods. If a vendor can't produce those five things within a week, you have learned what you needed to know. Confirm current plan details and trust documentation on each vendor's own site before you sign.

FAQ

Is SOC 2 or ISO 27001 more important for an AI visibility platform?

It depends on your buyers. US enterprise procurement usually asks for SOC 2 Type II first. EU and global buyers often prefer ISO 27001. The controls overlap by roughly 60 to 70%, so either is a good sign, and holding both is better.

Is a DPA required if the vendor only tracks public AI answers?

If no personal data is processed, maybe not. But the moment you connect analytics, Search Console, CDN logs or user accounts, personal data is likely involved and Article 28 applies. Sign the DPA anyway, since account data alone usually qualifies.

What if a vendor says SOC 2 is in progress?

Ask for the target date, the auditor's name, and whether the audit is Type I or Type II. Then compensate: tighter access scopes, shorter retention, and no CMS write access until the report arrives.

Does ISO 27001 certification mean the vendor is GDPR compliant?

No. ISO 27001 covers information security management. GDPR also covers lawful basis, data subject rights, transfers and more. Certification helps with Article 32, nothing more.