The sunbeam Show: Episode 2 ft. Matt Bowers – Does Programmatic SEO still work in the AI era?
Thenuka Karunaratne
CEO & Co-Founder, sunbeam
22 min read
Matt Bowers is the co-founder of Pike, a growth agency that works with startups on SEO and organic growth. He is the former Head of SEO at Zapier and Director of SEO at Zillow, two of the companies that pioneered the playbook for programmatic SEO. He started at Zillow as an intern in 2010 and joined Zapier in 2019.
In episode 2 of The sunbeam Show, Thenuka speaks with Matt about whether programmatic SEO still works in the AI era, what separates product-led pages from AI-generated content, and how the playbook has changed. They also cover getting buy-in at companies like Zillow and Zapier, how SEO teams are changing with coding agents, advice for early-stage startups, whether SEO is still a viable career, the rise of agent discovery, and the two gates Matt uses for AI-assisted content.
This transcript has been lightly edited for clarity and readability.
There’s a narrative right now that programmatic SEO is dead, partly because it’s being conflated with AI-generated content. You built two of the biggest programmatic SEO initiatives at any company. What’s your take?
Matt: I don’t buy into it, but I think it’s getting a bad name because of all the slop that’s happening. From my perspective, programmatic SEO is still the same as it has always been, and it should stay that way. AI is just a tool to make it better. You can use AI to interpret data into sentences. Great. But if you’re just asking AI to write content for you, it’s an accelerated content machine, and you can’t call that programmatic.
A good programmatic SEO program ideally has the product as the landing page. If you think about Zillow, when you search for “San Francisco homes for sale,” you land on a map and a list of homes. It’s just the product. You can filter and search. We put a pretty URL on it, but it’s just an alias for some filters, like the location, the home type, or the number of bedrooms.
Zapier is similar. It’s basically showing you what you’re capable of doing in the product. Press one button and just do it. What a lot of the AI programmatic SEO plays are doing right now is saying, “We want a page for every city in the country.” Then you tell Claude to make that, and Claude writes an essay about every city in the country. They publish it and call it programmatic SEO, and it just ends up being bad content. You could have written bad content before. It would have been expensive, but you could have done it. It wouldn’t have been called programmatic SEO, because a human would have been involved, and it would have taken months or years instead of one day.
So there’s definitely a good version and a bad version.
Would you say the prerequisite for calling something programmatic is a unique, differentiated data set that’s core to the product you’re exposing publicly?
Matt: It’s almost essential. If it’s not differentiated, you can still get away with it if you have a great product. Homes around the country, for example, are not that differentiated, to be fair. Every Zillow, Redfin, and Realtor.com has the same data set. It’s about making it great for the user to land after a really meticulous Google query.
You can put a lot of dumb stuff into Google and get a great result on Redfin, Realtor.com, or Zillow, because everybody at those companies has done a great job setting up landing pages that are built into the product.
With Zapier, we did have a differentiated data product. When I worked there, I think we had 7,000 apps and all of the actions you could take with them. No one had anything close to it, and that made it really easy to do SEO around use cases and around the triggers and actions we could support.
Now it’s even deeper, because the AI capabilities of Zapier mean you can do practically anything. They have templates, and you can abstract out of user data into use cases and make landing pages for those, with one button into the product. So it helps when you have that differentiated opportunity, for sure.
How has the programmatic playbook changed before and after AI? For the companies doing it well, what is the adaptation playbook for this era of search, if there is one?
Matt: I’ve thought a lot about this over the last year, because clients are asking how they can do it faster. Like I said earlier, AI is a tool. It can interpret data into human-readable content, and it’s a way to ship faster. But I don’t think it has changed the playbook in general.
The playbook is: Do we have data that we can turn into scaled landing pages? If we’re only making 100 landing pages, we should probably just use writers or handcraft them. If we’re scaling to thousands or millions, like Zapier, you can’t do that. So that’s another prerequisite for programmatic: Is it necessary? Do you have data that’s big enough that you need lots and lots of pages?
Then there’s the ability to connect those landing pages to the product, so they’re very product-led and the bounce rate is low.
The way I see Google typically deciding whether a page is useful is this: When we sent traffic there, did people leave and go back to Google to do something else? The team at Zillow has been hyper-focused on that since probably 2016 or 2017, and got rewarded a lot for it.
Cross-linking is still important, but so is bounce rate. We ran all of our A/B tests internally. So part of the playbook is: Can we make something that actually meets the user’s need and is engaging, so they can continue their journey within our website? At the very least, can we answer their question fast enough that they get what they need and don’t go back to Google and ask someone else?
The playbook still has a ton of technical SEO, site speed, crawl, all that stuff. It’s the same as ever. AI is transforming a lot of things, but I don’t think it’s transforming what ends up on the page. It’s transforming how you got to that point, potentially, because engineers can move faster, and PMs can get better data straight from the source if they’re not writing their own SQL. You can learn what’s in there and plan how you’re going to approach it.
Do you think programmatic SEO is faring better or worse in the AI search era compared to editorial content, given that more discovery is happening through ChatGPT and Claude?
Matt: I don’t know if I have a data point to back up my opinion on this, but I have a feeling. Programmatic plays are usually data-backed. Someone did research or collected the data and put it together, so it’s trustworthy. Schema is usually on the page, and it’s organized nicely in these programmatic setups, whether it’s hotel search with prices or all of the real estate we’ve been talking about.
For editorial content, I feel like it’s getting less and less trustworthy. It’s just kind of eroding. I’ve had clients in the past who said, “We want to build an article about some topic, like the top 10 real estate portals or the top 10 real estate tools for agents.” Claude just reads the other 10 and rewrites them, and the data or the statements get slightly twisted. It’s like a game of telephone. I don’t think that really helps the internet.
Then is that information trustworthy? It slowly erodes. It’s more of a feeling, but it seems like it is. If I were ChatGPT, I would prefer the programmatic approach. But I’m not sure ChatGPT is robust enough to have gotten to that point with its own index. I think they’re still living vicariously through all of the indexes they purchase. They were using Google for a while, and I’m not sure if they still are. I think Google tried to block them, but they have ways.
Is anyone doing a particularly great job with programmatic SEO in the last couple of years? The canonical examples are Zillow, Zapier, and Coinbase. Have you seen anyone innovate on the playbook in a way you admire?
Matt: I’m a little biased, but my business partner, Derek, worked at Figma, and about a year and a half ago they put together an incredible font library and made it programmatic. You can sort and filter by serif, not serif, whatever. If you’re trying to pick a font on the internet, it’s by far the best way to do it.
If you look at the success of that programmatic play on Ahrefs or Semrush, it’s awesome to see. More recent than that, I don’t know if I have a good one. It’s happening all the time. If you get it to work, you usually don’t talk about it. When I was at Zillow, we didn’t go to conferences. All our secrets were ours. So I’m sure they’re out there, and I’m just not paying attention, or people are just keeping it quiet and collecting their visits.
What was the internal game like, from seeing the potential as the SEO person to getting everyone to buy in, shipping, and then iterating in a loop?
Matt: When it’s going well, it’s really fun. Zillow was remarkably bought into SEO as the engine that made the company go, from the early days. I started as an intern at Zillow in 2010. There were only 200 people at the company, and half of them were in sales. But everybody was bought into SEO, from the CEO down, because Trulia was kicking our butts. The only reason we were losing was SEO. Trulia had done something, and we had no idea what it was.
As an intern, my boss told me, “You’re working on this stuff. We’re going to figure it out.” From that point on, any time we had an insight, whether it was learning from what Trulia was up to, hiring consultants to give us ideas, or research I did, as long as there was a document with a well-reasoned rationale, everybody was bought in from an existential standpoint.
At a certain point that fades, and it turns into just work and not “live or die on this project.” Then you build relationships internally. Back in the day, we worked in person, and it was about building those relationships to explain why SEO was so important to my team and why it would be important to theirs. If we could get this to work, we could collaborate.
At Zillow, the SEO team had four engineers at the most at any given time, and an engineering manager, plus six or seven on the SEO team. Those engineers were all working full time on SEO across the whole stack.
For many pieces of it, though, we would work with other teams. For example, the home details page, where your apartment or house has its own dedicated page on Zillow, has its own team of about six on the engineering, design, and PM side. We would bring a plan to them explaining why we needed something changed for SEO, then work with their PM and designer to implement it. Sometimes their engineers would work on it.
Getting buy-in from the top is very useful, so that it’s actually doable. Getting buy-in from below requires a lot of data and reporting, which we also spent a ton of time on. At Zillow, we had the best of both worlds. We had lots of bandwidth to be really robust about why things matter, cause and effect, what we’d tried that hadn’t worked, and what had worked.
In the end, you do all the work and go to the meetings to get your work prioritized on other teams. I think that’s pretty universal across all big-ish companies. You need someone else’s help, whether it’s SEO or not. It’s just part of the work of being a PM or a director.
Was it the same at Zapier? Was the CEO always bought in from the beginning?
Matt: Yes. I joined Zapier in 2019. The SEO program had begun in about 2012, and Zapier started in 2011. When I started, we tripled the size of organic traffic from 2019 to 2024, but it had already grown quite a bit. We were growing on a large base, because it was already a big deal, a household name among entrepreneurs.
The CTO, Bryan, had built the original landing pages back when there were about seven people at the company. So the execs were bought in because they saw it get built really up close, when it was a tiny startup. The reputation Zapier has as an SEO-first company is all true. Now it’s about 800 people. I think it was about 800 when I was there, and I assume it’s still about the same. If you put together a brief on why something is important, it’s possible to get it done pretty rapidly.
I’ve been blessed to work at places where the buy-in came easy. At Zillow, we showed initial results that were so good that the second layer of buy-in was really easy. At Zapier, I arrived and they hired me because they were already bought in. I had been doing SEO for seven or eight years at that point.
But there are many companies where there has been no SEO up to this point. It usually takes some kind of jolt to the system to get it activated. It could be pointing at a competitor that’s killing you, like we did at Zillow with Trulia, or paid ads getting too expensive and budgets running out. It doesn’t matter. You have to find that wedge internally to get the bandwidth to go fight the good fight.
How is the structure of SEO teams changing with coding agents? You mentioned Zillow had an engineering team of four. What does a mature SEO organization look like today?
Matt: I’m sure it varies by company, but it’s so easy to get things done now. If you learn Git and have access to Claude, it can be pretty dangerous. If I were running things inside of Zillow, I would still be excited to have four engineers with access to AI tooling, and I would expect us to ship way faster and experiment more. At a startup, I would be empowering my SEO team to actually ship to production. Let’s not cut any corners and let’s not break the site, but let’s move fast and find out what works. On the programmatic side, I would not want the SEO team publishing willy-nilly on the blog and saying, “Hey, we shipped 400 articles last month.” That’s not the use of AI I’m looking for.
But if you have a great data set and the SEO team wants to turn it into visuals and product, I think it’s a lot easier to move really fast, ship a bunch of stuff, and find out it didn’t work than it is to meticulously plan everything perfectly and then find out it didn’t work anyway.
So we might as well just go.
What’s your advice for very early-stage companies, pre-seed, seed, and sometimes Series A, that are thinking about SEO? When should they move forward and be early adopters versus holding off until their domain authority is higher?
Matt: They have the disadvantage of lower domain authority. They’re smaller and less resourced, but primarily their domain authority is just a lot lower. I have one specific client in mind that I’ve been working with for a year with exactly this conundrum. They’re in the kids’ activities industry, things like summer camps and stuff to do with your kids around town. Columbus Day was last week, or yesterday, whenever it was. You can use their website to find activities for your kids when school’s out and you’re at work.
It has sort, filter, age, and location, which is perfect for programmatic SEO. You can integrate it right into the product and make the product the landing page, which is my core principle for programmatic. But their domain rating is about 19 if you’re using Ahrefs.
The way I’m advising them is to make sure the foundation is in place so that as the product is built and grown, it continues to be SEO friendly. We’re going to train the engineers, of which there are just a handful, on what they need to look out for so they can self-direct in some ways. We’re going to set up the core principles, like the title tag formulas, the H1s, and the way the filters interact with the URLs.
Then what I’m telling my client is that they have to build a brand. We need people to go to the website in Chrome so that Chrome can feed the algorithm. We need people to go to Google and search for their brand name. We need people to use the website and recommend it to their friends so that it makes it onto the so-called mom blogs.
I also think affiliate is really important these days, way more than before. Affiliate has always been a good way to build your business if you can afford to give up percentages. But now you can get an army of affiliates saying nice things about you by giving them a cut, and you only have to pay them when they find business for you.
I have a story about that, if you’re interested. When I was at Zapier, around when ChatGPT first came out, we started tracking how it talked about us, the way you can now do with AirOps, Peec, or any of these prompt-tracking tools. I set up prompt tracking by hiring offshore workers on Upwork.
Every Monday, my team would go on the free version of ChatGPT, type in my prompts, copy the responses into an Excel spreadsheet, and do some analysis for me on whether the responses about Zapier were positive, negative, or neutral. We weren’t querying the API. We were typing the prompts into ChatGPT and tracking who came back.
We were also identifying which competitors were promoted. At the time, there was Integromat, which is now Make, a big brand that spent a ton of money, out of the Czech Republic. And there was a company called Pabbly that nobody’s ever heard of, but they do workflow automation too. I went back through and figured out that early ChatGPT was regurgitating, “You can use Pabbly to do that automation.” If you want to connect Meta ads to get a text message, a classic Zapier use case, it would say, “Pabbly would be good for that.” I was like, who’s Pabbly?
It turns out they had an amazing affiliate program with big payouts, and tons of affiliates on all these junk websites were just saying, “Hey, you can use Pabbly.” Zapier never had an affiliate program. We just had marketing. In the end, ChatGPT was just bought into it.
For this client, I’m telling them we should be talking about affiliates because it’s an easy way to get more content on the internet, which may help on that side. But the reason I want it is that I want them to build a brand. If they think the affiliate angle is financially feasible, let’s do it.
There’s also a weird side effect I’ve witnessed, where sometimes the models get confused. I also think it happened with early training data. At least in my consciousness, it felt like the first couple of big models OpenAI released were trained on a pristine, no-AI version of the web. Now you can’t do that. Most of the web is GPT-3.5 or GPT-4 writing it, and then they’re training GPT-6 on GPT-4’s content.
For someone thinking about SEO as a career path today, given how much is doable with agents and how much has changed over the last three or four years, is becoming an SEO professional still viable for someone who is new and wants to learn?
Matt: Definitely a viable career path, and more important than ever. The breaking-in part is tough for a lot of industries right now. It’s a conundrum. I was reading something in the Wall Street Journal the other day about lower entry-level job openings than there have been historically, and many more seniors. The job market is skewed. There are just as many jobs as before, but they’re all senior, which is tough. How do you become a senior if you never get to start?
As an example, take the kids’ camps client I was talking about. If they were to hire someone to work on SEO full time, I would say they could probably do all the stuff I was doing in 2010 in one day a week. Now they have four days a week of productivity available to them, and the question is what we do with it.
So if you’re trying to get into entry-level, full-time SEO, either you’re going to be incredibly productive and there’s an awful lot to do at your company, so you can learn a lot more, a lot faster, or you’re probably going to end up being more of a generalist. You take care of SEO on Monday. On Tuesday, you do social. Then you work on the ads team.
That feels like a much more standard expectation going forward. I would imagine that if you’re hiring someone onto the marketing team, you’re expecting them to wear a lot of hats as an entry-level person and find out what they’re great at. Eventually, they get senior in one of those things.
But I do feel like it’s a weird time to expect to be entry-level and doing just one thing all the time, unless you’re joining a huge team like Zillow that says, “We’re going to train you, you’re only going to be doing SEO, and we have a full plate for you right away.” Maybe, but it is tricky.
The discovery experience is being condensed into one set of surfaces, so there may be more opportunity in the future, even though the execution work feels more compressed right now. How do you see it?
Matt: I agree. I’ve never thought about YouTube as much in my whole career as I have this year. Gemini and AI Overviews use YouTube constantly as a citation. Before, it would have been a nice-to-have, and video production is expensive and takes time to make great.
Same with Reddit. It would always be nice if the Reddit results said something nice about our brand, but we weren’t going to have the social media team or the customer service team working on Reddit. It wasn’t a high enough priority for the SEO team.
But because ChatGPT is going to regurgitate what’s on Reddit, potentially, it matters now. It’s less than before. It used to be a crazy number of citations, but it’s still out there. So how do we deal with all those off-site things as an arm of SEO? It’s a big part of the game now that it wasn’t even two or three years ago.
Whether we call it AEO or whatever, I call it LLMO. The AI search world is a lot on your own site, but it’s also a lot off your own site, which we used to just call link building. Now we don’t need the link. The link is nice to have, but we’ll take the positive sentiment citation as well. So that is evolving quickly.
What would you say is the most overrated SEO and AI search tactic, and what is the most underrated?
Matt: The most underrated AI search tactic is probably writing great content on your own website. It’s really useful. You can use it to build authority and to build your personal brand and your company’s brand. If you do it right, it has all these side effects. It explains why customers love you. Now when ChatGPT reads it, there it is. It says right there why customers love sunbeam.
It’s kind of table stakes. Nobody talks about it because it’s not sexy, but it’s pretty good stuff. Our agency spends an awful lot of time and energy on that for our clients, with great success. We show it to them in the monthly reports we give our clients, so they can see it’s been worth it.
For overrated, I have yet to see a successful, long-term AI content push. It was fun in 2024. I even had a couple clients where we shot to a million visits with a quick win, but it just doesn’t seem to last.
I’m not sure where it lands on the overrated and underrated spectrum, but I think you have to think really hard about what you’re doing with AI. If tokens are turning into text and going on your website, you’ve got to think really hard about it. I think you should be coding with those tokens.
What do you see from the teams that are using AI content in a way that works? What is the strategy that embraces what’s been unlocked without going down the route of being undifferentiated and non-performing?
Matt: There are two really easy gates that we put up. The content that’s going live should either have a subject matter expert involved in the creation, even if it’s just a voice memo, or it should have data backing the article or whatever content piece you’re working on.
There are exceptions to that rule, where you don’t need either of those things and it can still be great. But if you have one of those, or both, it typically slows down your pace if you need a subject matter expert. Now you’re pacing more appropriately, because you’re getting human thought in the loop.
The content is going to offer something unique, either sourced from cool data that you collected or from someone who has deep knowledge of that domain. I would almost be comfortable saying, “Hey, you can one-shot with Claude if you gave it a two-minute voice memo. Great. Feel free to publish that.” Please read it and edit it, but hopefully you get my point. That would be better than most things that are going out with just AI these days.
The sunbeam Show features conversations with SEO and growth leaders who go deep on the one thing they know better than anyone else, recorded over morning chai. Hosted by Thenuka Karunaratne, co-founder and CEO of sunbeam.