Skip to content

Using GPT-4 with website visitor data: an example

I thought I’d give you an example today of how I’ve been investigating the use of AI (GPT-4) for SEO, website visitor analysis and content generation ideas. This one is unashamedly based on How to Combine ChatGPT and GA4 to Find Topics for Search and Social Media on the Orbit Media Studios blog.

Note that this uses the GPT-4 ‘Code Interpreter’, which requires a subscription to ChatGPT. However, even if you don’t have a subscription and can’t try it with your own data, I think you’ll find it interesting.

The data I’ve used in this investigation comes from a blog written by a local councillor where I live. She is always looking for new topics, and wants to understand what subjects resonate with local residents.

What I’ve done is to provide some very basic Google Analytics visitor data from the blog – just the titles of the pages and the visitor numbers/engagement. I’ve requested GPT-4 to analyse the table, which is what the Code Interpreter module is required for. From this, I’ve asked GPT-4 to recommend new topics that might generate good traffic and engagement. I’ve then asked it to analyse the titles to see if there’s a link between these and the traffic to the articles.

Note that this is just a start on the exploration, and based on some very elementary data. We can do much better, by providing more data and asking more questions. Just the page titles and visitor numbers really doesn’t give the AI much to analyse.

Some of its basic analysis would be clear to any of us, but even so, the speed in which GPT-4 can produce a report is a real time-saver. After that, we should be able to start asking questions that will give us some real insight that might otherwise require significant expertise. I hope it gives you some real food for thought. Happy experimenting.

  • User (Chris)

I’m giving you some Google Analytics data about traffic to a blog written by a local councillor in Cambridge, England, aimed at local residents. It shows the performance of all of the content in the last year. Can you answer questions about this dataset?

  • ChatGPT

Of course! Please provide the Google Analytics data, and I’ll be happy to help answer any questions you have about it. If the data is in a file format, you can upload it directly. If it’s text-based, you can paste it here. Once I have the data, you can proceed with your questions.

  • User (Chris)

(Uploads data-export.csv)

  • ChatGPT

Thank you for providing the data. I’ll start by loading and examining the first few rows to get an overview of its structure.

Finished working (Show work)

The dataset contains information on various pages or articles from the blog, including:

  1. Page title and screen name: The title of the article or the name of the screen.
  2. Users: The number of unique users who viewed the page.
  3. Sessions: The number of sessions during which the page was viewed.
  4. Average engagement time: The average time (presumably in seconds) users spent on the page.
  5. Engagement rate: The proportion of users who engaged with the page.

Now that I have an understanding of the data structure, please let me know the questions you’d like answered regarding this dataset.

  • User (Chris)

You are a digital marketing analyst and an expert at making data-driven recommendations. Combine the data from the columns to recommend new topics that would likely have good traffic and engagement. Organize the topics into two groups: those that would perform well in social vs. organic search.

  • ChatGPT

To recommend new topics that would likely generate good traffic and engagement, we should analyze the existing topics’ performance. Here’s a general approach:

  1. Identify High-Performing Topics:
    • Traffic: We’ll use the “Users” and “Sessions” columns to gauge popularity. High values here indicate topics that draw users in.
    • Engagement: “Average engagement time” and “Engagement rate” will inform us about user retention and interest. High values here suggest content that keeps users’ attention.
  2. Categorize by Acquisition Channel:
    • Social: Content that is shareable, trendy, or evokes strong emotions typically performs well on social media platforms.
    • Organic Search: Evergreen content or content that answers common questions tends to perform well in organic search results.

For the categorization by acquisition channel, without direct data on where the traffic is coming from (i.e., social media vs. organic search), we can make educated guesses based on the topic’s nature and engagement metrics:

  • Social: Topics with a high engagement rate might be more shareable and therefore more suitable for social platforms. This is because shareable content often evokes reactions, leading to higher engagement.
  • Organic Search: Topics with a longer average engagement time might be more informational and detailed, making them suitable for organic search. Users searching for specific information are more likely to spend time reading in-depth content.

Let’s begin by identifying the top-performing topics based on the metrics mentioned above.

Finished working (Show work)

Based on the analysis, here are the top topics categorized for performance on different platforms:

Recommended for Social:

  1. All That Glitters
  2. Cambridge’s dysfunctional development trajectory
  3. Scooters, art and planning permission: all on one day!
  4. The Two-Speed City
  5. Talkin’ bout a revolution

Recommended for Organic Search:

  1. In the end, it’s the wrong question
  2. Profit and Loss
  3. Scooters, art and planning permission: all on one day!
  4. Setting our future direction
  5. The Two-Speed City

Note that some topics, like “The Two-Speed City” and “Scooters, art and planning permission: all on one day!”, appear in both categories. This suggests they are versatile topics with potential for high performance across both social media and organic search.

For new topics:

  • For social, consider topics that are similar in nature to “All That Glitters” or “Talkin’ bout a revolution”, which could be trendy, evocative, or shareable.
  • For organic search, delve deeper into topics like “In the end, it’s the wrong question” or “Setting our future direction”, which seem to provide comprehensive or unique information.

Remember, while this analysis provides an educated guess on content categorization, the true determination of content suitability for a platform would benefit from additional data such as specific traffic sources, backlink analysis, and social shares.

  • User (Chris)

Suggest 10 new topics that are not in this dataset. Divide them into two groups: good for social and good for search.

  • ChatGPT

Certainly! I’ll use the insights derived from the analysis, as well as general knowledge of content that typically performs well on social media versus organic search.

Recommended for Social:

Social media platforms thrive on content that is shareable, timely, trendy, or evokes strong emotions. Based on the top-performing topics from the dataset and general trends, here are some suggestions:

  1. “The Changing Face of Cambridge: A Photographic Journey”: A visual piece that showcases the transformation of Cambridge over the years.
  2. “Cambridge’s Hidden Gems: Unexplored Corners”: Highlighting lesser-known spots in Cambridge for locals to explore.
  3. “The Art and Soul of Cambridge: Street Artists to Watch”: A feature on local street artists and their work.
  4. “Cambridge Festivals: A Cultural Dive”: Highlighting upcoming local festivals, their significance, and what to expect.
  5. “Voices of Cambridge: Inspiring Stories from Locals”: Short interviews or features of local personalities or everyday heroes.

Recommended for Organic Search:

Organic search tends to favor evergreen content or content that answers specific questions and provides detailed information. Here are some topic suggestions:

  1. “A Comprehensive Guide to Cambridge’s Historical Sites”: An in-depth look into the history and significance of various landmarks in Cambridge.
  2. “Cambridge’s Architectural Marvels: A Deep Dive”: Detailed articles on unique architectural sites in Cambridge.
  3. “Public Transport in Cambridge: Everything You Need to Know”: A guide on the public transportation system, tips, routes, and etiquette.
  4. “The Evolution of Cambridge’s Urban Landscape”: Discussing the development trajectory, planning decisions, and their impact.
  5. “Cambridge’s Education Institutions: A Comparative Analysis”: Offering insights into various educational institutions, their history, and what they offer.

These topics are just starting points. Depending on the blog’s audience, the writer’s expertise, and current events or trends in Cambridge, they can be adjusted or expanded upon.

  • User (Chris)

Titles are very important when it comes to generating inbound traffic. Can you detect any connections between the titles of the more successful pages on the site?

  • ChatGPT

Absolutely! The title of a page plays a crucial role in attracting readers. To detect connections or patterns among the titles of the more successful pages, we’ll examine various aspects:

  1. Length of Title: Determine if there’s a specific length (in terms of words or characters) that tends to be more successful.
  2. Common Words or Phrases: Identify any recurring words or phrases in the titles of successful pages.
  3. Tone or Sentiment: Check if the successful titles tend to be positive, negative, neutral, or curious in nature.
  4. Structure: Examine if there’s a particular structure (e.g., questions, lists, or statements) that appears more frequently.

Let’s start by analyzing the titles of the top-performing pages (let’s say the top 20% based on the performance score) to see if we can detect any patterns.

Finished working (Show work)

Based on the analysis of the titles from the top-performing pages, here are some observations:

  1. Length of Title:
    • Average word count: Approximately 4 words.
    • Average character count: Approximately 27 characters.
    • Titles tend to be concise and to the point.
  2. Common Words or Phrases:
    • The word “the” appears frequently, which isn’t surprising as it’s a common definite article.
    • Words like “it,” “and,” and “all” also appear frequently, but these are general words. The recurrence of these words doesn’t provide a specific theme or pattern.
    • The word “our” suggests a collective or community-focused approach in some titles.
  3. Structure:
    • There were no titles in the form of questions among the top-performing pages.
    • There were also no list-style titles (those that might start with a number, like “5 Ways to…”).

From this analysis, we can deduce that the successful titles on the blog tend to be short and direct. While we identified some common words, they are quite generic. This suggests that the success of the content might be more attributed to the relevance and quality of the content itself, rather than specific keywords or phrasings in the title.

For future content, sticking to concise titles and focusing on the quality and relevance of the content might be a good approach. However, experimenting with various title structures, such as posing questions or creating list-style titles, could also be considered to see if they resonate with the audience.