AdLens

Framing the Problem

What is the problem you are trying to solve? Who does it affect?

Social media has become one of the biggest sources of product recommendations, but users often struggle to tell whether an influencer is genuinely recommending a product or promoting it through a commercial partnership. Although advertising regulations require sponsored content to be clearly disclosed, compliance remains inconsistent. According to the UK's advertising regulator, only 57% of influencer advertisements complied with disclosure rules, 9% attempted disclosure using unclear or misleading wording, and 34% contained no disclosure at all. More than half of the problematic advertisements were in the fashion and travel sectors.

This lack of transparency affects everyone. Consumers may make purchasing decisions without understanding commercial incentives, content creators risk losing audience trust when partnerships are not clearly communicated, and brands can suffer reputational damage from inconsistent or inadequate sponsorship disclosure.

Idea Explanation

What is your idea? How does it fix the problem?

Our idea is to build a tool that increases transparency in influencer marketing by helping users identify whether an Instagram post is likely to be sponsored.

Users simply paste an Instagram post caption or the embedded code, and our system analyses the post using a machine learning model trained on over 1.6 million labelled Instagram posts. Instead of relying solely on hashtags like #ad or #sponsored, the model considers signals such as caption content found in sponsored posts. The tool then provides a sponsorship prediction along with a confidence score and the key factors that influenced the result. 

Our solution helps users make more informed decisions by identifying sponsorship signals and highlighting potentially undisclosed commercial content. Rather than replacing the user's judgement, we provide a transparent assessment that explains why a post may have been flagged.

Just as Wise makes hidden fees transparent, our project makes hidden commercial relationships on social media more transparent.

Implementation

How do all the pieces fit together? Does your frontend make requests to your backend? Where does your database fit in?

Our system consists of three main components: the frontend, the backend, and the machine learning model.

The frontend allows users to paste an Instagram post caption/ embedded code and sends the request to the backend through an API.

The backend processes the request, extracts the relevant post information (such as the caption and metadata), performs the same feature engineering used during training, and passes these features to the trained machine learning model.

The machine learning model was trained offline using a labelled Instagram dataset containing over 1.6 million posts. Once trained, the model is saved and loaded by the backend to make predictions on new posts.

The model then returns:

Sponsored / Not Sponsored

Confidence score

Key transparency signals

Finally, the backend sends this result back to the frontend, where it is displayed as a simple transparency report for the user.

Training (Offline)

Instagram Dataset

        ↓

Feature Engineering

        ↓

Train ML Model

        ↓

Saved Model (.pkl)

────────────────────────

Prediction (Online)

Instagram post caption/ embedded code

        ↓

Frontend

        ↓

Backend API

        ↓

Feature Extraction

        ↓

Load Trained Model

        ↓

Prediction

        ↓

Frontend Results

Challenges

What did you struggle with? How did you overcome it?

One challenge we identified is how to deploy a transparency tool fairly and sustainably. Our project benefits consumers by helping them identify potentially sponsored content, but it also affects influencers, brands, and social media platforms. Incorrect predictions could damage reputations, while overly strict detection could discourage legitimate advertising.

To address this, we designed our system to support transparency rather than make accusations. Instead of declaring that a creator is dishonest, our tool highlights sponsorship signals and provides a confidence score, allowing users to make their own informed judgement.

Our project creates value for multiple stakeholders, not just consumers.

• Consumers gain greater transparency and can make more informed purchasing decisions.

• Brands and agencies can use the same detection engine as a private compliance tool to review influencer content before or during campaigns, reducing the risk of unclear sponsorship disclosure.

• Content creators can check whether their posts meet disclosure standards before publishing, helping them build trust with their audience.

• Regulators and platforms benefit from technology that encourages clearer advertising practices and improves transparency across social media.

Rather than exposing or penalising brands, our goal is to help all parties identify potential disclosure issues early and promote responsible, transparent influencer marketing.

Looking ahead, maintaining and improving the system would require ongoing support. Potential funding could come from government consumer protection initiatives, regulatory organisations, or social media platforms, as these stakeholders all have an interest in improving transparency and promoting responsible advertising.

Accomplishments

What did you learn? What did you accomplish?

Through this project, we learned how to develop an end-to-end machine learning application, from data preprocessing and feature engineering to model integration and frontend deployment. We also gained experience working with a large real-world social media dataset and designing a solution that balances technical performance with ethical considerations.

We successfully built a tool that analyses Instagram posts for sponsorship signals and presents the results in a clear, explainable way. Rather than simply making a prediction, our system promotes transparency by helping users understand why a post may have been flagged.

Next Steps

What are the next steps for your project? How can you improve it?

Currently, our model is primarily trained using captions and hashtags. We chose this approach because recent changes to Meta's platform policies and API restrictions limit access to richer post metadata for public Instagram content.

In the future, if additional data becomes available through authorised APIs or partnerships, we could improve the model by incorporating features such as images, engagement metrics, tagged brands, and creator profile information. This would enable a more robust multi-modal model and improve prediction accuracy.

Next, we would improve the project by expanding beyond Instagram to other platforms such as YouTube, TikTok, and X, where sponsorship disclosure is also often unclear.

We would also improve the model by focusing on specific sectors such as makeup, skincare, food, fitness, and technology, because sponsored content can look very different across industries.