If you had looked at my ChatGPT history three months ago, a lot of it would have looked familiar: content ideas, rewriting something, researching a topic, fixing a paragraph.

That is no longer how I use AI.

Over the last 90 days, I have slowly moved AI much deeper into the day-to-day operations of the businesses I work on.

I use it while working on Sendwo, our WhatsApp engagement platform. I use it for HowToBuySaaS, where we are building a SaaS discovery and marketplace business. I use it while working on my personal website, sales processes, SEO, WordPress, lead management and the dozens of small decisions that eat up a founder’s day.

Some days, ChatGPT is helping me figure out which sales lead deserves attention first.

Twenty minutes later, I may be using the same conversation to debug a WordPress category template.

Then I might be looking at Search Console data, rewriting a landing page, reviewing pricing, preparing a proposal or deciding what I should stop working on.

This experiment has changed my opinion about AI.

I don’t think the interesting question anymore is, “Can AI write content?”

We answered that a long time ago.

The more useful question is:

How much of the operating work around a small business can AI help one person handle?

After spending roughly three months testing that question in my own businesses, I have a much better answer.

Quite a lot.

But only if you know where to use it.

I stopped treating ChatGPT like a search box

My early use of AI was fairly normal.

Ask a question. Get an answer.

Ask for a blog outline. Get an outline.

Ask it to improve an email. Copy the result.

Useful, but not particularly interesting.

The bigger change happened when I stopped starting a new conversation for every small problem and started giving AI the actual context of the business.

What are we selling?

How is the product priced?

What happened with this lead yesterday?

What objection did the customer raise?

Which SEO pages have impressions but poor positions?

Which page did we update last week?

What is currently blocking this project?

Once that context builds up, the interaction changes.

Instead of asking:

“Write a follow-up message.”

I can ask:

“This lead signed up for the free plan, contacted us through live chat, wants the official WhatsApp API and has already booked a demo. What should I do next?”

That produces a very different answer.

The quality of AI output, at least in my experience, depends heavily on how much reality you give it.

AI has become surprisingly useful for SaaS sales operations

Sales follow-up is one of those jobs that sounds simple until you have 30 conversations in different stages.

One person asked about pricing.

Another completed a demo.

Someone signed up but never activated the product.

A company wants an API.

Another company wants bulk messaging but has a privacy concern.

Someone said “interested” four days ago and then disappeared.

The problem isn’t writing a WhatsApp message. I can write a WhatsApp message.

The problem is keeping track of what happened, remembering the context and deciding which conversation deserves attention today.

For Sendwo, I started using AI alongside a simple CRM workflow.

For every lead, I can keep information such as the use case, last action, current concern, next follow-up, expected revenue path and what I should avoid doing.

Then I can look at the day’s leads and ask AI to help me prioritize them.

That has been far more useful than asking it to “generate sales copy.”

A lead who has booked a demo should obviously not receive the same message as someone who created a free account and vanished.

Someone evaluating an official WhatsApp API setup shouldn’t receive the same pitch as a small business looking for a basic marketing tool.

AI is good at maintaining those distinctions if the information is already there.

It is also useful for something founders are terrible at: remembering every follow-up.

I have had days where a fresh lead feels exciting, so I naturally want to chase it, while an older lead that is much closer to revenue gets ignored.

Giving AI the pipeline makes that bias easier to spot.

It does not close the sale for me. It helps me avoid losing sales because my attention went somewhere else.

That difference matters.

Where AI has helped me most with SEO

I have worked in digital marketing for years, so I wasn’t looking for AI to teach me what a title tag is.

What I wanted was help processing more information without spending half the day doing it.

HowToBuySaaS is a good example.

The website has SaaS product pages, category pages, comparison content, reviews and informational articles. There are a lot of possible things to work on.

That creates a familiar SEO problem:

What should I update first?

Looking at Search Console one URL at a time is possible. Doing it repeatedly across a large site becomes tedious.

So I started bringing actual page and query data into my AI workflow.

One page may already have tens of thousands of impressions and an average position somewhere on the first page.

Another may have almost no impressions.

The first one probably deserves attention before I spend six hours creating something completely new.

This sounds obvious when written down.

It is much harder to follow consistently when you are managing a business.

We have used this process while reviewing pages around products such as FaceCheck ID, Janitor AI and NeverCap AI, and while rebuilding SaaS category pages covering CRM, SEO software, WhatsApp marketing, email marketing, project management and other categories.

The AI isn’t making the final SEO decision.

It helps me inspect the evidence.

For example:

Why might a page get impressions but very few clicks?

Does the page actually answer the search intent quickly?

Is pricing information missing?

Would comparison content make sense?

Are users probably looking for a free option?

Does the page need clearer screenshots?

Are there obvious internal links we haven’t created?

Are FAQs answering real buying questions or just filling space?

This is one area where I think people sometimes use AI backwards.

They ask AI to generate 100 article ideas and then start publishing them.

I’d rather start with the data I already have and ask AI to help me understand where the existing opportunity is.

AI is much better when it is reacting to evidence than when it is inventing a marketing strategy from nothing.

Coding has been the most unexpected part

A fair amount of my recent work has involved WordPress, custom HTML, CSS, PHP snippets and Elementor.

And anyone who has worked seriously with WordPress knows how these stories go.

You make one change.

Everything looks fine.

Then you open the mobile version.

Something has moved six kilometres to the right.

Or you fix a category template and suddenly discover that the category description no longer appears.

That actually happened while we were rebuilding HowToBuySaaS category pages.

The design worked, but the long description we were adding from the WordPress backend stopped appearing on the frontend.

So the conversation became less about “write me some code” and more like debugging with another person sitting beside me.

Here is the existing code.

Here is what it currently produces.

Here is a screenshot.

Here is where the content is stored in WordPress.

Don’t redesign everything.

Don’t remove the parts that already work.

Fix this particular behaviour.

Then check mobile.

This is where AI coding becomes useful for people running small businesses.

You don’t always need AI to build a completely new application.

Sometimes you need it to help you fix the annoying 40-line problem that has already consumed two hours of your day.

There is a catch.

I have also seen AI confidently provide code that is technically valid and still wrong for my implementation.

That is why screenshots, existing code and exact error behaviour matter.

“Build a responsive category page” is a weak instruction.

“The desktop version is correct, the bottom category description is missing, here is the PHP currently running through Code Snippets, don’t change the product grid, and mobile content should collapse” is a useful instruction.

Specificity wins.

I now use AI before making some business decisions

This part is harder to quantify, but it may be the part I value most.

Founders make hundreds of tiny decisions.

Should this be $29 or $49?

Should I follow this prospect again?

Does this feature deserve another week?

Should this page exist?

Is this landing page confusing?

Should we give a discount?

Should this lead be moved to low priority?

Do I need another product feature or do I need more customers?

It is easy to confuse thinking with progress.

Sometimes I already know what I want to do, but I need something to attack the idea before I spend money or time on it.

AI is useful for that.

I can explain the situation and ask it to find the weakness in my plan.

I can tell it that a customer rejected a price and ask whether discounting is actually sensible.

I can give it the workload for the day and ask which task is closest to revenue.

I can explain that five projects all feel urgent and force myself to rank them.

This is not sophisticated technology.

It is structured thinking.

But structured thinking is valuable when you are switching between product, sales, marketing, customer support and finance all day.

My best AI conversations often don’t produce something I can copy and paste.

They help me decide.

Content creation still matters, but my workflow has changed

I obviously still use AI for writing.

The difference is that I don’t find completely AI-generated content very interesting anymore.

It tends to become generic quickly.

You can usually feel when an article has been produced from a title and nothing else.

There are plenty of correct sentences, but nothing happened.

Nobody tested anything.

Nobody got something wrong.

Nobody changed their mind.

Nobody lost a customer.

Nobody had to fix the same WordPress template three times.

That is why I increasingly start with what actually happened.

For HowToBuySaaS, that may be something we discovered while working on SaaS SEO.

For Sendwo, it could come from a customer conversation.

For SnehilTalks, like this article, it can come from the way my own work has changed.

AI then helps me organize the thought, question weak parts, improve clarity and turn scattered notes into something readable.

That workflow feels much more useful to me than:

“Write a 2,000-word article about artificial intelligence in SaaS.”

There is already enough of that content on the internet.

AI is very good at remembering the boring stuff

This sounds like a small benefit. It isn’t.

A big part of operating a business is repetitive context.

What price did we quote?

Which plan was this person considering?

What did we promise to change?

When should we follow up?

Which pages have already been updated?

What was the next step on that project?

You don’t want to spend your best mental energy remembering administrative detail.

When a long-running AI conversation has the relevant context, I can return to something and continue from where I left it.

That makes AI feel less like a tool I occasionally open and more like an operating layer sitting beside the business.

There are limits to this, of course. Important company records should still live in proper systems. I don’t believe your entire business should exist inside a chatbot conversation.

But as an interface for working through those records, it is extremely convenient.

What has not worked

There have been enough failures during these 90 days that I no longer accept a good-looking AI answer as evidence that the answer is good.

One recurring problem is confidence.

AI can be wrong in a very calm voice.

I have seen it misunderstand how a WordPress implementation works, assume functionality that wasn’t there, produce copy with facts I never provided and recommend strategies that sound sensible until you compare them with the actual business.

Another problem is overbuilding.

Ask AI for a CRM and it may happily design something that looks like Salesforce had a baby with an ERP system.

Meanwhile, you have 20 leads.

The same thing happens with marketing strategies.

You ask how to grow a SaaS company and suddenly you have 14 channels, a community strategy, webinars, partnerships, influencer marketing, product-led growth, an affiliate program and a podcast.

Wonderful.

Who is doing all of that on Tuesday?

The answer is nobody.

AI doesn’t naturally feel your resource constraints. You have to give those constraints to it.

If I have limited time, I say so.

If I don’t want another sales call, that matters.

If revenue this month matters more than a six-month branding exercise, that matters.

If I am the person who has to execute the recommendation, that definitely matters.

The quality of the answer improves when AI understands what you cannot do.

I don’t trust AI with facts just because it sounds certain

This deserves its own section because it has changed how I work.

When the answer depends on something current, I verify it.

Pricing changes.

Software interfaces change.

Google changes things.

Meta changes things.

Policies change.

Companies shut products down.

Job openings disappear.

Search results move.

An AI answer based on old information can be perfectly written and completely useless.

So I have become much stricter about separating two types of work.

There is reasoning based on information I provide.

And there is information that needs to be checked against the outside world.

I trust the first category much more.

If I provide my own Search Console numbers and ask for help analysing them, there is something concrete to work with.

If I ask, “What is the latest policy for X?” without checking the source, I am taking unnecessary risk.

AI works better for me when I treat evidence as an input, not something the model is expected to magically know.

The prompts matter less than people think

I have never been particularly interested in collecting “100 ChatGPT prompts for entrepreneurs.”

Real work doesn’t happen like that.

I rarely use a beautifully constructed prompt.

My conversations are messy.

“Look at this.”

“This is wrong.”

“Don’t change that section.”

“The customer replied with this.”

“We already tried that.”

“Here is the screenshot.”

“Why did this break?”

“Forget the previous pricing. This is the latest one.”

That probably looks terrible in a prompt engineering course.

It works because the conversation contains context.

I think people sometimes spend too much time trying to discover a magical sentence that will produce the perfect AI answer.

The more useful skill is learning how to provide information.

Show the numbers.

Paste the customer response.

Share the current code.

Explain what you’ve already tried.

State the constraint.

Correct the AI when it misunderstands something.

Keep the conversation attached to reality.

That has done more for my results than any prompt template.

AI works best when the feedback loop is short

I have noticed another pattern.

AI is most useful when I can test its output quickly.

If it gives me a CSS fix, I can paste it and see whether the mobile page breaks.

If it suggests a follow-up message, I can send it and see whether the prospect responds.

If it recommends changing the opening of a landing page, I can implement it.

If it finds a weak section in an article, I can rewrite it immediately.

The result comes back into the conversation, and the next answer becomes more informed.

This is much better than asking AI to create a giant six-month strategy that nobody revisits.

My preferred cycle now looks something like this:

Problem -> context -> AI input -> human decision -> execution -> result -> next decision

The important part is the result.

Without that, you’re just having an intelligent-sounding conversation.

Has AI replaced a team for me?

No.

And I think “AI replaces employees” is often the wrong framing.

What has changed is the amount of work I can get through before I need another person involved.

There are tasks where previously I might have waited for a developer, spent much longer investigating something myself or simply postponed the work.

Now I can move further on my own.

That doesn’t mean the specialist has no value.

It means I can approach the specialist with a narrower problem.

The same applies to marketing.

AI can help analyse a page. It cannot make a weak product worth buying.

It can help prepare a sales message. It cannot create customer trust on its own.

It can help write code. It doesn’t understand every consequence of deploying that code to a production website.

It can help me challenge a decision. I am still responsible for making it.

For a small team or founder-led company, that extra leverage is significant.

The biggest gain has been reducing the distance between idea and execution

This is probably the most important thing I learned.

Ideas are cheap. Most founders already have too many.

The delay happens after the idea.

You think a page should be redesigned, but first you need the code.

You know you should follow up with old leads, but first you need to organize them.

You want to analyse SEO data, but first someone needs to go through it.

You want to test a different offer, but first somebody needs to write the page.

Every “first” creates friction.

AI removes some of that friction.

Not all of it.

But enough that an idea can become something testable much faster.

That has affected how I work more than any individual AI feature.

What I would tell another SaaS founder starting today

Don’t start by trying to “implement AI across the company.”

Pick an annoying part of your own work.

Something you repeatedly postpone.

Sales follow-ups.

Customer research.

SEO analysis.

Documentation.

A small coding problem.

Reviewing support conversations.

Writing product FAQs.

Preparing meeting notes.

Give AI the real material involved in that job and see whether it can shorten the path to completion.

Then keep using it on the same workflow.

The second week will probably be more useful than the first because the context improves.

And don’t hide the messy parts from it.

If your budget is low, say the budget is low.

If you have no team, say you have no team.

If you hate doing sales calls, include that constraint.

If you only have four hours available, don’t ask for a strategy designed for a 15-person marketing department.

AI doesn’t need a polished version of your business.

It needs the actual version.

Where I am after 90 days

I started using ChatGPT like most people did.

Ask something. Get something.

Today, it sits much closer to the way I operate.

Some conversations are about revenue.

Some are code.

Some are SEO.

Some are product.

Some are basically me arguing with an AI until I figure out what I actually think.

And yes, sometimes the answer is bad.

Sometimes I tell it the same thing three times.

Sometimes it fixes one piece of code and breaks another.

Sometimes it writes something so polished that I immediately know I don’t want to publish it.

That is part of using the tool properly.

The goal isn’t to accept more AI output.

The goal is to get more useful work done.

After 90 days, that’s the part I am convinced about.

AI has not removed the need to understand my business.

It has made understanding my business even more important.

Because once execution becomes easier, the quality of the decision behind that execution matters a lot more.

And for me, that is where AI for SaaS operations has started becoming genuinely useful.

Frequently Asked Questions

How can SaaS founders use AI in day-to-day operations?

AI can help with lead prioritization, sales follow-ups, SEO analysis, content editing, customer research, documentation, basic coding, website troubleshooting and reviewing business decisions. It works best when it has access to the real context behind the task.

Can ChatGPT manage SaaS operations?

ChatGPT can assist with many parts of SaaS operations, but I wouldn’t treat it as an autonomous business manager. I use it as a working layer around my own data, decisions and execution. Final responsibility still sits with me.

Is ChatGPT useful for SaaS sales?

Yes, especially for maintaining lead context, preparing follow-ups, analysing objections and deciding which opportunities deserve attention. I find this more useful than simply asking it to generate cold sales messages.

Can AI help with SaaS SEO?

It can be very useful when combined with real SEO data. I use AI to inspect Search Console data, review existing pages, identify content gaps, analyse search intent, plan internal links and decide which pages deserve attention first.

Can AI build or fix a SaaS website?

AI can help write and debug HTML, CSS, JavaScript, PHP and other code, but the output still needs testing. I have found it particularly useful for solving smaller WordPress and frontend problems when I can provide the current code, screenshots and exact behaviour I want to change.

What should you avoid outsourcing completely to AI?

I would be cautious with business-critical facts, financial decisions, legal matters, production code and anything based on information that may have recently changed. AI can help investigate or structure the problem, but verification still matters.

Does using AI actually save time for a SaaS founder?

For me, yes, but the biggest saving hasn’t come from generating text faster. It comes from reducing context switching, analysing information more quickly, remembering ongoing work and shortening the gap between deciding to do something and actually doing it.