AI transparency

How We Use AI: Data, Permissions and Human Oversight

AI agents are useful for monitoring, analysis and first drafts. They also make mistakes. This page explains what our agents do, what data they use, what needs a person’s approval and what AI cannot promise.

Last updated: 16 November 2026

In one paragraph

Marketing2u AI agents help identify opportunities, prioritise issues and automate selected, low-risk tasks. Important changes stay under your control. Agents only read the data you connect, receive the lowest level of access their job needs, and hand over to a person when a situation is sensitive, unclear or high-stakes.

What our AI agents do

  • Monitoring: checking Search Console, Analytics, ad accounts or a website on a schedule and flagging changes.
  • Analysis: summarising performance data, grouping search queries, spotting unusual changes and comparing periods.
  • Recommendations: suggesting what to fix or test next, with the evidence behind each suggestion.
  • Drafting: first drafts of reports, FAQs, page copy, ad variations, social captions and replies to common enquiries.
  • Routine actions: where you have agreed in advance — for example sending an acknowledgement to a new enquiry or logging a lead in a CRM.

What automates and what needs approval

Default handling by action type
Type of actionDefault handling
Reading connected data, monitoring, internal reports and alertsAutomated
Answering routine questions from information you approvedAutomated, with hand-over to staff
Publishing website content, blog posts or social postsHuman approval required
Changing a live website (titles, redirects, code, settings)Human approval required
Changing ad budgets, bids or launching campaignsHuman approval required
Quoting prices, offering discounts or making commitments to customersHuman approval required
Replying publicly to reviews or complaintsDrafted by AI, approved by a person
Deleting data or contentNot delegated to agents

These are our defaults. The exact rules are set in your app settings or, for managed setups, written into the proposal. They can be made stricter, never looser without your written agreement.

For a practical way to match review effort to task risk and reversibility, read our human oversight guide for AI-assisted marketing.

What data agents use

Agents only use data from sources you connect or provide, such as Google Search Console, Google Analytics 4, Google Ads, Meta, your website, your CRM, a shared inbox, or documents you give us (service lists, prices, FAQs and policies). Competitor monitoring uses publicly available information only.

We ask clients not to connect sources containing sensitive personal data (for example NRIC numbers, medical details or financial account data) unless the workflow genuinely requires it and the handling has been agreed in writing.

Access and permissions

  • Separate user accounts, not shared passwords
  • Read-only access wherever the job allows
  • Each agent gets only the access its role needs
  • You can revoke access at any time from each platform’s own settings
  • Access is removed when a project ends

AI models and third-party services

Our agents use large language models and other AI services provided by third-party technology companies, together with the APIs of the platforms you connect (such as Google and Meta). Data passed to these services is limited to what the task needs. If you need to know which providers are used for your project — for example for your own PDPA assessment — ask us and we will tell you before anything is connected.

Personal data and the PDPA

Where agents handle personal data — for example names and phone numbers in enquiries — both your business and Marketing2u have responsibilities under Singapore’s Personal Data Protection Act 2012. We design workflows to collect only what is needed and to keep it within the systems you control where possible. The Personal Data Protection Commission (PDPC) publishes guidance for organisations, including on the use of AI. Our own handling of website enquiries is described in our Privacy Policy.

Limitations you should know about

  • AI can be wrong. Language models can state incorrect things confidently. That is why outputs that reach customers or the public are reviewed.
  • Recommendations are only as good as the data. Broken conversion tracking or incomplete CRM records lead to poor recommendations.
  • No outcome guarantees. AI cannot guarantee rankings, traffic, leads or sales. Google, Meta and your customers decide those outcomes.
  • Platforms change. Integrations can break when Google, Meta or other providers change their systems. We monitor connections, but gaps can happen.
  • Not every task should be automated. Work that needs judgement, empathy, negotiation or professional accountability stays with people.

How recommendations are generated

A typical recommendation combines three things: data from your connected systems (for example a page’s impressions and click-through rate in Search Console), rules and checklists configured for your business (for example “service pages must link to the contact page”), and a language model that summarises the finding in plain English. Each recommendation should show the evidence it is based on so you can check it.

AI and data — FAQ

Can the AI modify my website automatically?

Only if you explicitly agree to it for specific, low-risk changes. By default, website changes are drafted by the agent and applied after a person approves them.

What happens before the AI publishes something?

The draft is placed in an approval queue or sent to the named approver. Nothing public is published until that person approves it, unless you have agreed otherwise in writing for a specific task.

Can it connect to Google Search Console?

Yes. Search Console lets site owners add users with restricted permissions, which is how we connect. Google describes the permission levels in its Search Console help documentation.

Do AI agents replace people?

Not entirely. Agents handle monitoring, analysis and drafting well. Strategy, creative judgement and accountability still need people, which is why every agent works within approval rules and important actions are approved by you or your team.

How is my data protected?

Through least-privilege access, separate accounts, limiting which data is connected, human approval for important actions and the ability to revoke access at any time. No system is risk-free, so we explain what each agent can access before it is connected.

Questions about AI and your data?

Ask us before you connect anything. We will explain exactly what an agent would access and do.

Trust & accountability

Advice you can verify. Actions you control.

Marketing2u is an owner-run Singapore business. We explain what our AI agents can and cannot do, keep people responsible for important decisions, label illustrative workflows clearly, and do not guarantee rankings, leads or revenue.

Who is responsible

Published by Marketing2u under Alvin’s direction. Read the real-world experience and ownership disclosures behind our work.

Who runs Marketing2u →

How work is checked

Our process covers evidence, human review, approval gates, AI assistance, corrections and content updates.

Our methodology · Editorial standards

Data and AI transparency

Connections and permissions depend on the configured tools. Important publishing, spending and customer commitments remain subject to agreed human approval.

How we use AI and data →

Found an error or need a claim clarified? Email info@marketingagency.com.sg with the page URL.