seoaiagent.ai

Right now, I am building seoaiagent.ai, an AI local SEO intern for agencies.

The simplest way I think about it is this: when an agency hires a real intern, they do not give them full control on the first day.

They give them a checklist, examples of good work, client notes, and access to a few tools. They review what the intern produces, correct mistakes, and slowly trust them with more.

I want to build the AI version of that.

Not a tool that gives an agency another audit and leaves the work to them. Not a chatbot that acts as if it knows more than the SEO specialist. And not a fully autonomous employee that changes a client’s website or Google Business Profile without anyone checking.

The product should learn how the agency already works, follow its SOPs, help with repeated delivery work, and prepare things for the team to review.

At the moment, I am starting with local SEO work such as competitor research, Google Business Profile audits and monitoring, keyword opportunities, drafts, and updating the tools an agency already uses.

I am also working with one local SEO agency with a team of 10 people and talking with other agencies to build the first version. It took me a while to arrive at this version.

It started with Vedic AstroGPT losing traffic

I entered SEO through Pramish, my college senior and mentor. He had built Vedic AstroGPT, an AI astrology product where a meaningful part of the users came through Google. SEO had not been worked on consistently, and the traffic had started falling. That was my entry into the problem.

I did not know much about SEO then. I understood that search traffic mattered, but I had never thought deeply about what produced it, why a page ranked, how rankings changed, or how much repeated work went into maintaining them. Pramish suggested that there might be an opportunity to build something around this.

The first customer in our head was someone exactly like him: a technical founder who had built a useful product, had some search traffic, and did not want to spend their own time doing SEO every week. The original idea was not specifically about local SEO. It was broader: could we build an AI employee that handled SEO for founders?

Before building that, we first needed to understand the work. So we began doing it ourselves. For roughly three months, we studied SEO, experimented on real websites, and built a rough internal system around the work we were learning to do manually. It helped with research, content preparation, tracking, and organizing the next actions. The system was broad because our understanding of the problem was broad. We thought the user was a founder like Pramish. Then we tried to find more of them.

We started with people who had launched on Product Hunt

Our first customer-acquisition idea was simple. People launching products on Product Hunt were building software. Many of them wanted growth. Some of them would probably need SEO. So we went through Product Hunt launches, filtered the founders who looked relevant, found their contact information, and sent around 200 cold emails.

Only two replied. It was one of my first real lessons in the difference between someone matching an ICP on paper and someone actually wanting the thing you are selling. The founders looked like the right customers to us, but most of them had not told us that SEO was a current problem. We were contacting them because they had launched a product, not because they were actively looking for SEO help.

After that, we changed the way we searched for customers. Instead of starting with a list of people who looked right, we started looking for people already talking about the problem. We searched through X, Reddit, Facebook groups, and other online communities where founders were asking questions about traffic, content, rankings, and SEO. That worked better because the need was already visible.

Our first customer came from X

Our first paying customer was the founder of WhispriNote, an AI note-taking product. I found him through X. The money itself was not what made it important. He was not a friend, a relative, or someone introduced by a mentor. He was a person outside our network who had seen enough value in what we were offering to pay for it.

It was the first indication that this might be more than an internal experiment. But serving a SaaS founder also made one limitation clear: we were still learning SEO while trying to sell it. The work was often open-ended. A SaaS company could need technical SEO, content, links, landing pages, conversion improvements, or many other things. Each client could lead us into a completely different workflow.

So we began looking at businesses where the connection between search visibility and revenue was easier to understand.

Then we worked with an e-commerce company

One of our next clients was ONIN Infosys, an e-commerce business in Nepal selling laptops, computers, and electronics. This was a more serious engagement than the earlier trial, and it exposed us to a different kind of problem.

ONIN had a custom-built website. We could research the opportunity, identify what needed to change, and prepare the recommendation. But we could not always implement it ourselves. A seemingly small SEO fix could require help from their developers. We would explain the change, wait for it to be added, check whether it was implemented correctly, and then go back again if something was missing.

The work was not only:

Find an SEO problem and solve it.

It was also:

Understand the website, understand who controls it, communicate the change, wait for implementation, verify it, and keep following up.

That made me realize that the ability to automate SEO depended heavily on the client’s website, CMS, team, and internal workflow. The recommendation itself was often the easy part. Getting it executed was harder.

Local businesses made the problem more direct

After e-commerce, we began exploring local service businesses. The relationship between search visibility and business outcomes was much easier to understand there. These businesses did not need massive traffic. They needed the right nearby people to find them at the moment they were searching for a service.

We began reaching out across many categories of local service businesses. I tried almost every outreach method I could think of. I called businesses directly, sent cold emails, and contacted people through LinkedIn, X, Reddit, Facebook groups, and WhatsApp.

For some businesses, I recorded a short personalized Loom video. I would open their website or Google Business Profile, point out a few specific problems, explain what I thought they could improve, and send the video through WhatsApp.

Those videos worked much better than our generic cold emails. The owner could immediately see that the message was not copied and sent to hundreds of businesses. It was about their website, their profile, and the things they were currently missing. Across this period, we made around 300 cold calls. Some conversations ended immediately. Some people were curious but never replied again. Some agreed to meetings. And some became customers.

This changed how I understood SEO. Until then, it had been easy to think of SEO as rankings, keywords, pages, and reports. Once we were doing the work for real local businesses, it became much more operational.

We had to understand the services each business actually wanted to prioritize, which locations mattered, what type of customers they wanted, which competitors were relevant, what information was missing from their profiles, what their developers could change, and what their internal staff could realistically maintain.

Two businesses could both want to rank in the same city and still need completely different strategies. One might want volume. Another might want a more specific kind of customer. One might have a team member available to collect photos and reviews. Another might barely respond to messages. The business context changed what good SEO work meant.

Doing the work ourselves gave us access to all these details. It also showed us how much of the delivery was repeated. For every client, we had to collect information, inspect the business profile, look at competitors, check rankings, analyze pages, prepare content, follow up on changes, and explain the results. That repetition was what made the possibility of an agent interesting. But we still had the customer wrong.

We assumed the business owner should use the software

After working with local businesses, we thought we understood the opportunity. Business owners needed more customers from Google. They lacked time and SEO knowledge. So we assumed the answer was to give them software that made SEO easier.

We spent around two months building a self-serve product for local businesses.

It had dashboards, progress tracking, Google Business Profile workflows, AI-generated replies, post creation, review-related features, and other things that seemed useful based on the work we were doing.

Then we gave it to business owners, receptionists, and local teams. They did not use it much. At first, it was tempting to blame the onboarding, interface, or missing features. But the deeper problem was simpler. They did not want to operate an SEO product.

The pattern was clear across the businesses we worked with. They wanted more customers, more bookings, and better visibility. They were focused on running their business, serving customers, and managing daily operations. None of them woke up wanting a better SEO dashboard.

We had made the work easier to see, but we had not removed the need for someone to understand it, decide what mattered, and make sure it was completed.

That was exactly the responsibility many of these businesses wanted to hand to someone else.

It took us about two months to build the product and another month to properly accept that the user did not want it.

We tried to find owners who might want to do it themselves

Before giving up on the self-serve direction, I tried a narrower version of the same idea. Maybe the problem was not business owners in general. Maybe more technical business owners would be comfortable using the product themselves. Computer repair shops seemed like a reasonable segment. The owners already worked with technology, managed websites, and understood computers better than the average local business.

I collected repair shops from Nepal, Chennai, and Bengaluru.

For more than 50 of them, I recorded personalized videos and sent the videos through WhatsApp. I showed what was weak on their website or Google presence and explained how the product might help.

The videos received far more attention than cold email, and I got on calls with several owners. But even here, the same issue appeared. Being technically capable did not mean they wanted to spend their time operating an SEO tool. They were repairing computers, managing staff, dealing with customers, sourcing parts, and running their shops. SEO was important because it could bring them business. It was not important enough for them to want another job.

That was the point where the distinction became clear to me:

The person who benefits from a problem being solved is not always the right person to operate the software that solves it.

Local businesses had the problem. But agencies and SEO professionals were more likely to be the users.