Identifying search keywords for your startup with Open AI ChatGPT

Tools Used In This Recipe

An old saying in startups is, “it doesn’t matter if your product is 10x better than alternatives if nobody knows about it”. Search engines, specifically Google, are the starting point for a massive portion of peoples’ research and purchase journey, so if you can get your product in front of them during that process you can build a sustainable funnel of interested prospects.

Building that capability takes a lot of work, and for a startup it can be a full-time job (on top of the other full time jobs that come with being a founder). In late 2022, OpenAI released ChatGPT, a chat-based layer on top of their powerful AI Large Language Model (LLM). ChatGPT allows you to interact with their AI engine using natural language. So you can ask it simple math questions like ‘What is 732 X 43” or make more complex requests like “Create a sea shanty about the TV show Ted Lasso”.

As a startup founder, ChatGPT can solve a lot of your day-to-day challenges. You’ll still need to edit the output a bit, but it’s extremely good at solving the ‘blank page problem’ where you don’t even know how to start to solve a problem. In this Recipe we’ll use ChatGPT to generate a list of search keywords that are related to Startup Recipes’ customer base. This example should give you plenty of inspiration to help you target your own customers.

Getting started

To get started go to the OpenAI API home page and click Get Started to log in or set up an account if you don’t have one. Once you have logged in, navigate to the ChatGPT home page.

Here you’ll find a very simple interface; a large window that will show the interactions between you and ChatGPT, and a small text box in the bottom of the screen where you can enter your queries. Although ChatGPT uses a natural language interface and can generally understand your intent without any specific structure, OpenAI has an excellent article on tips for getting the best results that is well worth your time to read.

Since we are looking for a list of Google searches our target customer might make, we’ll start with the following prompt:

The first sentence, I want you to act as a marketer, helps ChatGPT narrow down its focus. Since it contains billions of text examples from different industries and functions, telling it to focus on marketing will help it ignore potentially unrelated responses. In the second sentence we ask for a list of searches for both informational and transactional intent, so we can start to build our SEO and paid search strategies. We also describe the product, using enough detail so ChatGPT has the context it needs to generate a response. In the last sentence we ask ChatGPT to limit its response to search terms that get a high volume of searches.

In a few seconds ChatGPT comes back with a list.

Notice that several of the transactional intent searches aren’t as relevant for a free library of information like Startup Recipes, but for a purchasable product they would make more sense. If it is applicable to your product you can ask ChatGPT to generate a list of Google searches for commercial intent. Notice that ChatGPT automatically uses the context from your previous prompt - you don’t have to tell it to do that.

Now that we have established a list of search terms we can start to focus more on the specific actions we need to take. Asking ChatGPT to generate a list of keywords we should optimize for in SEO gives us a good place to start.

We asked ChatGPT to generate search phrases that had a high volume of search activity. So let’s verify them from the source. In other Recipes we’re used the Google Ads keyword planner to estimate the size of a market, but it can also give us ranges of search queries in a month for a set of search terms. Google Ads keyword planner expects terms to be in a list separated by commas, so we could take the list it generated and edit to to the format that the keyword planner expects, or we could just ask ChatGPT to do that for us.

Open the Google Ads Keyword planner, click Get search volume and forecasts, and paste the list of keywords into the box. Then click Get started.

The results however are not encouraging. Few of the terms even have enough monthly searches to register in this tool.

So what happened? The information we entered seemed correct and they output ChatGPT gave us looked accurate, but it wouldn’t be useful to generate content to match these search terms if there’s nobody searching for them.

This scenario illustrates one of the potential weaknesses of AI tools like ChatGPT. They will always give you an answer, and the more powerful they are the better the answer will seem, but none of this is helpful if you give the tool a bad prompt. What should we have done instead? Although ChatGPT will accept input like “generate a list of high-ranking search terms” its language model is not actually connected to Google’s Search index. So all it can do is make a guess based on the data it’s been trained on, and these inferences don’t always work well. However the real issue is more fundamental. We asked ChatGPT to give us to help market a product, Startup Recipes, instead of asking ChatGPT to envision our target customer and try to guess what problems they would have. A common entrepreneurial mistake; focusing on your product instead of on your customer.

Let’s go back to ChatGPT and try a different approach. We’ll ask it to generate a persona for a typical startup founder (and ask it to ignore demographics because it’s very possible the OpenAI dataset contains a biased representation of the type of person who would be a startup founder), then ask it to list some common problems that person would face and Google searches they would use to help solve them.

ChatGPT provides an indented list starting with the persona, the list of problems, and for each problem a set of Google searches that they might use.

Let’s dig into one of the problems and suggested searches.

Converting these to a comma-separated list and going back to the Google Ads keyword planner shows more promising results.

While the highest-volume queries Google reports average more than a million searches per month, it’s still a good idea for an early stage startup to focus on lower-volume queries. These ‘long tail’ terms typically have lower competition and lower cost per click, and for an early stage company trying to get its first ten, hundred, or thousand customers even a term with 10,000 searches per month can have meaningful results.

Finding related queries

Once you have found a set of search terms that seem promising you can use that list to find related queries. Google’s search page includes a section called ‘people also search for’. It’s intended to help searchers adjust their queries if they are not finding good results, but it can be extremely helpful for marketers. Dashword has compiled a lot of this data and released it as a free tool on their site, so it can be a great place to check for additional search terms to add to your marketing campaign. Let’s take one of the keywords and test it in Dashword’s tool.

There are two terms here that weren’t part of our list but have decent search volume, so they’re definitely worth adding.

Wrapping up

ChatGPT and other generative AI tools have tremendous ability to support your startup’s marketing efforts. As we learned in this Recipe, they will sometimes confidently give you an answer that turns out to be unhelpful, so always try to validate it against primary sources (like we did with its suggestions for high-ranking search terms). But even with its limitations ChatGPT should be a key tool in your startup’s toolkit.

The contents of this Recipe are © Innovation Works, Inc. and are licensed under CC-BY-SA 4.0 . Contact us with questions or feedback, or to learn more about our structured program in Entrepreneurism based on Startup Recipes.