Quick answer
To ask a company to explain an algorithm’s decision about you, send a written subject access request under UK GDPR. Ask for a copy of your personal data, whether any automated decision-making or profiling was used, meaningful information about the logic involved and what it means for you. If a decision with a legal or similarly significant effect was made by a system alone, you can also ask for a human review and challenge it. Organisations usually have one month to reply. For feeds and recommendations, the quickest fix is often the platform’s own controls. If you are stuck, the ICO can help, and for serious disputes get qualified advice.
Key takeaways
- A subject access request (SAR) is free in most cases and covers data, profiling and automated decisions
- You can ask for meaningful information about the logic, not the company’s source code
- For significant decisions made by a system alone, you can ask for a human to review it
- Organisations normally have one month to respond, extendable for complex requests
- Wrong data about you can be corrected under the right to rectification
- For recommendation feeds, platform controls usually work faster than a formal request
- If the reply is poor, complain to the company first, then to the ICO
📘 Main article: This guide supports our full article, AI Interpretability: Why Goodfire’s Research Is Good News for Safety. Read it for why researchers now think AI decisions may become far easier to explain.
In this guide
- What can you ask a company about its algorithm?
- Which route fits your situation?
- How do you ask for an explanation, step by step?
- How do you correct what a recommendation feed thinks about you?
- What if the company does not answer properly?
- Which mistakes should you avoid?
- Frequently asked questions
- The bottom line
Most of us have had a decision made about us by a system we cannot see: a declined application, a frozen account or a feed that has decided who we are after one video.
In our article on AI interpretability, we cover Nate B Jones’s forecast that people will eventually be able to inspect and correct what an algorithm has learned about them. You do not have to wait for that. This guide answers one question: how do you ask a company to explain an algorithm’s decision about you today?
What can you ask a company about its algorithm?
You can ask what personal data they hold, how they used it and whether a system made or shaped the decision. In the UK, these rights come from UK GDPR and the Data Protection Act 2018.
| You can ask for | What it gets you |
|---|---|
| A copy of your personal data | The information the decision was based on, so you can spot errors |
| Whether automated decision-making or profiling was used | A clear yes or no on whether a system was involved |
| Meaningful information about the logic involved | A plain explanation of the main factors, not the code itself |
| The significance and likely consequences | What the decision means for you now and later |
| Correction of inaccurate data | Wrong details fixed under the right to rectification |
Be realistic about the logic part. Companies must give meaningful information, but they do not have to hand over trade secrets or a full model breakdown. As the main article explains, even the people who build many AI systems have struggled to explain individual decisions, which is part of why interpretability research matters.
Which route fits your situation?
Pick the route that matches what went wrong. A formal request suits serious decisions. A settings change suits a feed that has misread your tastes.
| Situation | Best first route |
|---|---|
| Declined loan, credit, insurance or job application | Subject access request plus a request for human review |
| Account suspended or flagged by an automated check | Ask the company’s support team for review, then send a SAR if needed |
| Wrong details in your profile | Request correction (rectification) in writing |
| Unwanted targeted marketing | Object to direct marketing, which includes profiling for it |
| Feed or recommendations that misread you | Use the platform’s own controls first |
⚠️ Warning: The rules on automated decisions were updated by the Data (Use and Access) Act 2025. The core safeguards, such as being able to challenge a significant decision and ask for human involvement, remain, but check the ICO’s guidance for the public for the current detail. For a dispute with money or a job at stake, get qualified legal advice.
How do you ask for an explanation, step by step?
Put the request in writing, be specific and keep a copy. There is no official form, and you do not need to mention the law by name, but it helps.
- Find the right contact. Look for a data protection officer or privacy email in the company’s privacy notice.
- Say it is a subject access request. Use those words in the subject line so it reaches the right team.
- Identify the decision. Give the date, reference number and what happened.
- List what you want. Your personal data, whether automated decision-making or profiling was used, meaningful information about the logic and the consequences for you.
- Ask for a human review if a system made a significant decision on its own, and say briefly why you think it is wrong.
- Note the deadline. Write down the date you sent it. One month is the normal limit, and they can extend it by up to two further months for complex requests if they tell you.
- Keep everything. Save emails, letters and screenshots in one folder.
A short template you can adapt:
“This is a subject access request. On [date], my [application/account] [reference] was [declined/suspended]. Please send me a copy of my personal data, confirm whether automated decision-making or profiling was used in this decision, and provide meaningful information about the logic involved and its consequences for me. If this decision was made without human involvement, I would like a person to review it.”
💡 Tip: Ask for the main factors in plain English. A question such as “Which pieces of information most affected this decision?” is easier to answer well than “Explain your algorithm”.
How do you correct what a recommendation feed thinks about you?
Use the platform’s own controls first, because they act straight away. Most large platforms let you tell them a recommendation is wrong, remove items from your history or reset what they show you.
- Mark items as not interested rather than just scrolling past them
- Remove specific items from your watch or search history instead of clearing everything
- Check ad and interest settings, where some platforms list the topics they think you like
- Be precise. Reject the one show or topic you dislike, not a whole category
This is exactly the frustration Nate describes: you dislike one show and the system decides you dislike a whole genre. Today’s tools only let you nudge. If you want to see what a platform has recorded about you, a subject access request works here too.
What if the company does not answer properly?
Complain to the company first, then to the Information Commissioner’s Office (ICO). The ICO is the UK regulator for data protection.
- Reply to the company, quote your original request and say what is missing.
- Allow a reasonable time for them to fix it.
- If they still do not respond properly, raise a concern with the ICO, including copies of your correspondence.
Which mistakes should you avoid?
- Asking for “the algorithm”. You will get a refusal. Ask for the main factors and your data instead.
- Phoning only. Without a written record, it is hard to show when you asked.
- Missing the decision reference. Vague requests take longer.
- Treating an AI-written reply as proof. A neat explanation is not evidence that it matches how the system actually worked, which is the gap interpretability research is trying to close.
- Giving up after one reply. A polite follow-up often gets a fuller answer.
Want the bigger picture? Read AI Interpretability: Why Goodfire’s Research Is Good News for Safety, and if you use AI agents at work, see How to Spot Reward Hacking When an AI Agent Says the Job Is Done.
Frequently asked questions
Does a subject access request cost anything?
Usually not. An organisation can only charge a reasonable fee or refuse in limited cases, for example when a request is clearly excessive or repeated.
Can a company refuse to explain its algorithm?
It can protect trade secrets and does not have to share code, but it should still give you meaningful information about the logic and the main factors behind a decision about you.
How long does a company have to reply?
Normally one month from receiving your request. It can extend that by up to two further months for complex requests, but it should tell you within the first month.
Can I ask for a person to review an automated decision?
Yes, for decisions with legal or similarly significant effects made without meaningful human involvement. Check the ICO’s current guidance, as the rules were updated in 2025.
Who do I complain to if a company ignores me?
Complain to the company first, then raise a concern with the Information Commissioner’s Office.
The bottom line
You already have a way to ask what an algorithm used and why: a clear, written subject access request, plus a request for human review when a significant decision was automated. It will not give you a perfect window into the model, but it gets you your data, the main factors and a route to challenge mistakes. For feeds, start with the platform’s own controls, and save formal requests for decisions that matter.
Related posts
- AI Interpretability: Why Goodfire’s Research Is Good News for Safety
- How to Spot Reward Hacking When an AI Agent Says the Job Is Done
- How to Classify Business Data Before Sharing It with AI
[…] How to Ask a Company to Explain an Algorithm’s Decision About You — what you can ask for under UK GDPR today, with a simple template you can adapt. […]