Use AI to learn about who your users are
Eli says: “Everyone’s been talking about using AI for SEO, and they’re focusing on the wrong aspect of it.
AI shouldn’t be used to create content; it should be used to understand your users and figure out who they are. You can then use AI to create some of that content, but the magic of AI is understanding the things that you wouldn’t already know about your users.”
Is there a particular AI that you’re using?
“I was using ChatGPT, then I moved over to Gemini. I tried Perplexity but it always introduces itself by repeating the prompt, and I find it really annoying.
I’ve just started using Claude, which I think is the best of all of them.”
How do you use AI to understand who your users are?
“Ask it questions. If I want to get some ideas on what to post on LinkedIn, I could say something like, ‘Imagine I’m Eli Schwartz, the author of Product-Led SEO. What should I post on LinkedIn?’, and it comes up with fantastic ideas. I’m not asking it to write the LinkedIn post for me, I’m asking it to give me ideas. I like doing that for persona research.
If I’m working with a retail client that sells makeup, I would say, ‘I am a retailer. I sell makeup. Tell me who my persona should be.’ Then, the AI will spit out ideas. There will be lots of useless stuff, and then it’ll have some interesting nuggets.
It’ll say, ‘Here’s Becky. She’s a buyer of this makeup. She’s X age, she lives in Y place.’ Then you can follow up and say, ‘What kind of content does Becky read? How much time does Becky spend on TikTok? How much time does Becky spend on search? What kind of computer does Becky have? Does Becky use a phone? Does Becky use a computer?’ These are really detailed things.
From there, you can really personify and get an image in your mind of what your user looks like. Then, you can start writing content. That’s the part that has always been missing in SEO. You go to a keyword research tool, put in a primary keyword like ‘makeup’, and it gives you things related to makeup, but they’re not really attached to that buyer/user.
I spent years at SurveyMonkey, so I fell in love with the concept of building surveys, asking users questions, and learning really interesting things. Of course, you could do that before AI, but AI allows you to ask a fake person and get ideas. Is it perfect? Who knows? You have to do a survey to match that up. However, it does give you a different direction to be going in.”
What data do you feed an AI and what prompts do you give it?
“When it comes to SEO, we look at a lot of useless data. Actually, the answer doesn’t necessarily change if you add more data, in a lot of areas of marketing.
With surveys, for example, there’s the concept of statistical significance. That is the idea of whether the data you get matches the real-life scenario, and the findings are probably not due to random chance. If you want to know who’s going to win an election, you don’t have to ask every single voter; you ask a very small percentage of voters, and then you get a margin of error. If your findings are statistically significant, you know that you’re probably representing the truth.
When it comes to surveying for marketing purposes, you may have a really wide margin of error. If you ask, ‘Do most users like vanilla or chocolate ice cream?’ and 60% say they like chocolate and 40% say they like vanilla, that’s not necessarily significant. If you input more data from another audience, you may get 65/35% or 55/45% but, even though you don’t know the precise number, it’s unlikely to be completely inaccurate and most people actually prefer vanilla.
A lot of marketing data is trying to get to the perfect answer. However, if you feed more information in – explaining who you are, how many customers you have, how much your products cost, etc. – you won’t necessarily get more information out. At a base level, you can say, ‘I sell affordable makeup’, and then you’ll get an idea of what kind of personas come back.
You can say, ‘I sell luxury watches.’ You don’t need to say, ‘I sell luxury watches that cost X amount in Y specific neighbourhood.’ The answers won’t necessarily change. Ask a general prompt and explain what you’re trying to do. Then, the information that you get back can give you some ideas. You’re not building an entire business off those personas.”
How do you build confidence that the AI understands who your user is?
“It’s complicated. Most marketers don’t really understand their personas when they do any sort of marketing, because they never really ask. When I get these personas from AI, I don’t know that they are perfect. I don’t know exactly which one’s going to spend more money or where I should do all of my marketing.
However, if I get three personas back, that’s three more personas than I had to begin with. It’s not going to tell you that persona A is better than persona B, which is better than persona C. It’s giving you three places where you can start writing personalised content.
The bigger question is, if you got back three personas, how many more are there? In reality, there probably aren’t that many.
I’ve tried this with travel products, and there aren’t that many different types of travellers. There’s the budget traveller, the luxury traveller, the family traveller, and then little things in between. Yes, there is the solo teenage traveller and the Gen Z traveller, but the nuances between all of them are not that different, from a marketing standpoint.
Then, you can feed the personas back. Once you get those personas, you can ask it to tell you more about each one.”
Can you test these by seeing how closely the AI’s persona matches your existing personas?
“Of course, but that’s more work. I’m trying to make this easy by using the AI ‘cheat’. You could do even more work through a user survey.
All the time, with clients and potential clients, when I ask them what their users look like, they say, ‘How am I supposed to know?’ It’s baffling to me. Just ask them.
If you want product-market fit, you should know what the market looks like. You can get there accidentally, but getting product-market fit is about understanding your users and creating for those users.”
What prompts can help you understand buyer pain points?
“Once you understand what a persona is, you can say, ‘How many websites does a person like this look at?’ It’s hard to know if it will give you the right answer unless you do a real survey, but it has collected information from many different places, like Reddit, where people have written about how many websites they visit.
You can ask those kinds of questions to dig deeper into who this persona is. What I like about it is you don’t have to wait for those survey responses and for real users to answer the questions. You’re getting an idea.
Again, there are accuracy concerns. You have to test it, and you shouldn’t spend your whole marketing budget on it. However, you can dive deeper into it, as if you’re talking to that person.
If you’re selling affordable makeup, you can ask it follow-up questions. How much do they spend? How many websites do you look at? What kind of phones does this person use? Do they use an iPhone or an Android? Do they use the latest version of the iPhone? That might tell you more information about how much budget they have, and how often they buy.
You’re almost asking a real customer. Again, it might hallucinate, but that is still so much more information than you would have had without doing this process.”
Do different AI models have roughly the same level of hallucinations?
“I feel like Gemini hallucinates more, but they all hallucinate. You can even force them into hallucinations, which is kind of funny.
This is a first step. So many marketers don’t ever take this step, and they don’t understand their customer at all. This is not new. I spent years at SurveyMonkey and, as part of their onboarding, every single employee had to create a SurveyMonkey account. Almost none of them used those accounts, so they didn’t understand why people used their service.
How many Google Ads reps have actually run a Google Ads campaign? How many of the product managers working on Google Analytics actively analyse a website with it – or is it just a product to them?
Trying to understand your customers is something that everyone can benefit from. If you don’t have access to your customers, AI is a start. Will it hallucinate? Of course. Is it perfectly accurate? No, but it’s better than nothing.”
How often should you be doing this as part of an overall strategy?
“I look at all marketing as iterative. Generally, everyone should update their marketing around the users they’re learning more information about. You wouldn’t necessarily continuously prompt because you’re probably not going to get new information, but you should update anything you know about your personas based on your real-world experience dealing with them.
When it comes to content roadmaps and content marketing, a lot of marketers create content that’s very much at the top of the funnel. It’s not written for any specific user; it’s just being written. Understanding a persona helps you understand your buyer’s journey and your user funnel.
SEO is the best fit at mid-funnel. At bottom-of-funnel, there’s not as much search, and that’s where paid marketing is more successful. Top-of-funnel is those broad keywords, and it’s okay if you want to generate traffic, but it doesn’t necessarily generate conversions. You can use personas there, but you’re just driving traffic. Also, a lot of top-of-funnel is going to end up going away, as AI is integrated into search.
By really understanding your specific users, you can better understand what mid-funnel means, because now you’re focused on the funnel. Early on in my career, I worked at an automotive website and our top-of-funnel was the word ‘cars’ – as if someone buying a car didn’t know what they wanted to do. We ranked on the word ‘cars’.
That concept has obviously gone far away. You need to know who your user is and create content for that user. It starts with knowing your user so that you know what the funnel is.”
If an SEO is struggling for time, what should they stop doing right now so they can spend more time doing what you suggest in 2025?
“SEOs spend too much time reporting, and they’re getting metrics that don’t necessarily move the needle. Knowing the granularity around click-through rate for a specific keyword or knowing that something has moved from position 8 to 7 doesn’t always help you make better decisions.
It’s difficult when your management or your client expects this level of reporting, but a lot of reporting is never looked at. If you want to take back time in your day, try to have those conversations and push back on the reporting.
I had a job where I spent 1-2 hours per day, including on weekends, building a Google Sheet and populating it with numbers from Google Analytics and Search Console. Then, by accident, I discovered that none of the people I sent this sheet to ever opened it. I went on vacation, and I forgot to update it, and no one said anything. That day, I just stopped doing the reporting, and no one ever said anything about it.
If you find that you’re spending time reporting and it’s not moving the needle, and no one is making decisions based on that information, you can cut that out.
At least have conversations with your clients and determine the metrics that are key to them. Understand how they’re using the data (or ‘accidentally’ stop giving them that data and see what happens). The issue is that, sometimes, SEOs want to look like they’re busy and adding a lot of value.
Find those other things. Do persona research to add value rather than sending useless reports that people might not look at.”
Eli Schwartz is the author of Product-Led SEO, and you can find him over at www.ProductLedSEO.com.