Today we’ll be exploring how AI agents are learning to use computers: navigating websites, automating complex data generation, and transforming B2B lead research.

1. Website Navigation with Computer-Using Agents

Computer-using agents, are capable of navigating and interacting with websites in the same way a human user would.

Computer-using Agents can:

• Click Buttons: Identify and interact with buttons, even when labels or styles vary.

• Fill In Forms: Populate text inputs, dropdowns, checkboxes based on the user’s intent.

• Apply Filters: Interact with UI components like date sliders, tag checkboxes, or dropdowns.

• Traverse Multi-Step Workflows: Handle sequential interactions (e.g. login -> filter -> download).

• Handle Gated Content: Access pages behind logins

This enables complex data-gathering tasks that were previously impossible.

Watch the video below to see it navigate a the website “CV-Library”.

Example: Extracting Leads from a Job Board like Indeed/Cv-Library.

The software carries out these steps:

1. Inputs User Intent: e.g. “Find all Senior Product Manager roles in fintech companies in London.”

2. Translates Intent into UI Actions:

   • Enter role: “Senior Product Manager”

   • Set location: “London”

   • Apply filters for industry or company names (e.g. “fintech”)

3. Navigates Result Pages:

   • Click on each job post or extract from the listings view

   • Scroll and paginate to collect multiple results

4. Extracts Data:

   • Parse job title, company name, location, and description

   • Ignore irrelevant fields or sponsored listings

   • Return structured JSON or tabular data

5. Repeats the Workflow Reliably:

   • Restart the process with different keywords or geographies

   • Monitor success/failure and retry intelligently

What Makes This Possible

This entire process is enabled by three key capabilities:

• Browser Control: Using tools like Callm, the agent can control the DOM, take screenshots, or access accessibility trees to understand the layout and interact with it as a human would.

• Action Planning: The LLM doesn’t just receive a static prompt. It breaks the end-goal into intermediate steps and decides which actions to take next (e.g., “click the blue ‘Search’ button after filling filters”). This is similar to how a human would think through a task.

• State Awareness: Agents track what stage they’re in (search page, results list, inside a detail page) and adapt their behavior accordingly. They know whether they’ve succeeded in extracting data or if they need to retry.
  

Why It Matters

This capability shifts LLMs from passive processors of input text to active researchers. In lead generation, it means:

• Gathering fresh, real-time data from the open web

• Extracting information from any website, even those without APIs

• Adapting to page layout changes and UI variations

• Operating continuously and autonomously

As a result, AI-driven lead generation moves to live, intelligent exploration of the web, something only humans could do until now.

Until next time,

— Louisamay Hanrahan

In that time I’ve advised teams on tools that automate workflows and increase revenue, I’ve watched which tools actually stick, which deliver measurable value, and which are still overhyped or unwanted to begin with.

In this newsletter I’m going to give you a high-level overview of the key AI-powered software categories that I’m currently advising SMB B2B sales teams to explore. In the coming issues, I’ll dive deeper into each one with examples, case studies, pros/cons etc.

Let’s get into it:

AI & Automation Software Categories Transforming SMB Inside Sales

🔍1. AI-Led Lead Generation & Data Enrichment

What this enables: Personalised lead data generation aligned to your ICP, with live data enrichment.

Before AI, lead generation was largely one-size-fits-all. Most tools couldn’t automate leads for specific SMB niches or adapt to how teams actually define a good customer.

What surprised me most is how much AI has changed this. Tools now can provide personalised, daily up-to-date data. Two products I recommend are CALLM, which focuses on personalised lead data automation for niche SMB targets, and Clay, which supports teams that need high-volume, continuously refreshed lead enrichment for highly visible markets.

📞 2. AI Voice Cold Calling

What this supports: lead qualification and meeting generation

This is an area I initially underestimated. A client wanted to experiment with AI voice agents for outbound sales. We deployed AI-powered voice agents for cold calling and lead qualification, and they instantly booked meetings for our clients sales team.

Since then, I’ve noticed adoption across other SMB sales teams.

I assumed voice agents would be a natural fit for customer support teams in SMBs. However, the cost per call makes these tools difficult to justify for support use cases. Deploying them within the sales function, by contrast, changes the ROI equation.

Today, voice agents are best positioned as openers rather than full sales replacements. I don’t see them fully replacing enterprise sales teams. However, for companies that target smaller customers, they’re already taking bites out of the process.

As these agents become more capable, they’ll handle larger portions of the sales motion more effectively. This is an area I’m watching closely to see when voice agents reach the threshold where adoption accelerates rapidly due to improved capability and reliability.

If you are setting up your own voice agents I strongly advise using Retell over Vapi.

📧 3. Personal LLMs for Personalised Email Outreach

What this enables: Higher reply rates, improved deliverability, reduced manual sequence management.

Smaller SMB sales teams aren’t investing here yet. Instead, uptake is strongest among teams targeting higher-value customers and larger organisations, where the economics of deeper personalisation make sense.

One product example is Perlon AI, which offers highly personalised outbound email automation but typically starts at a four-figure monthly investment. For teams selling higher-ticket B2B products or enterprise-leaning offerings, this can be justified. For lower-ACV SMB sales motions, it’s often harder to make the numbers work.

🗒️ 5. AI Note-Takers & CRM Automation

What this enables: Reduced admin time and happier teams!

Tools that transcribe calls, summarise key points, and automatically update CRM fields, reducing manual admin for reps.

Example: Attio — call transcription and CRM updates built into the platform.

What stood out to me is how much value auto-updating the CRM adds. Salespeople are hired for their ability to engage customers, not for form-filling. Automating this reduces friction, improves data quality, and leads to happier, more effective sales teams.

I use Zoom and Otter’s note takers. Zoom is great because you can screen record.

🧾 6. Pre-Call Research & Prospect Briefing Agents

What this enables: faster call preparation and a safety net when there’s limited time for pre-call research.

Some teams use tools like SalesEcho, which delivers quick AI-generated pre-call briefings, summarising key details about a contact’s role, company signals, and recent activity, so reps have context before the conversation even begins.

You can use n8n and other agentic platforms but they can be tricky to set up.

🗣️ 8. Conversation Intelligence & Call Analytics

What this enables: insight-driven coaching, better rep training, higher conversion.

Tools that analyse call recordings to identify patterns, from objection trends to talk/listen ratios and messaging that consistently performs.

What this enables:

• Coaching based on real conversations, not anecdotes

• Faster identification of what messaging works

• More consistent performance across the team

Example: Gong, which analyses sales calls to surface trends, risks, and winning behaviours that managers can use to coach more effectively.

📅 9. Automate Your Inbox

For automating your email inbox, there are several great options. Tools like Fixer.ai can organise your inbox and draft replies. You can also use Gmail’s AI polish to clean up rough emails and auto-personalise pre-written templates, or dictate your email into ChatGPT and let it write it up for you (I tend to use this for longer emails, or ones where I’m trying to figure out the structure). There’s a lot that works well here.

📑 10. Slide Deck Automation

What this enables: Quick polished presentations

I’m a big fan of Gamma and other AI proposal-generation tools. They’re incredibly useful for quickly spinning up presentations for potential customers. You can also use ChatGPT to generate proposal copy, which speeds the process up significantly.

And That’s it!

These are the tools I’m seeing teams actually use. I’ll break down a few of them above in more detail in future posts. I’ll go into more detail on some of the tools above in future newsletters, particularly those I’ve implemented directly within client workflows. I’ll cover what’s worked, what hasn’t, how teams are actually using them, and where I think each tool is best suited in terms of team size, function, and use case.

If you have tools you love, that you’re testing, use cases you want me to evaluate, or questions on implementation, hit reply and let me know!

Until next time,

— Louisamay Hanrahan

Cloudflare declared **Content Independence Day** on July 1. From now on:

• **AI crawlers are blocked by default** for all Cloudflare users.

• The only way AI bots can access your content is if **you explicitly allow them**, or you **charge** for access.

This isn't a gradual rollout—it’s a **full stop** unless you take action.

🧠 Why This Matters

For years, AI crawlers roamed the web freely — scooping up your blog posts, articles, product descriptions, and documentation to train large language models. Whether you liked it or not, your content was part of the AI revolution.

But now, the rules are changing — and fast.

Cloudflare is introducing **Pay‑Per‑Crawl**, a new system that lets you:

• ✅ **Allow** AI crawlers (as before),

• - 💵 **Charge** them per request using HTTP 402 (yes, like a toll gate for your content),

• - 🚫 Or **block** them entirely.

And here’s the kicker: **if you do nothing**, Cloudflare now assumes you want to block.

This is the **first time** a major internet infrastructure provider has given you the power to *bill AI companies for crawling your site* — or shut them out entirely.

💣 Default Blocking

Imagine this: You publish an amazing blog post. But now, OpenAI’s GPTBot can’t read it. It won’t show up in summaries. It won’t be cited by Perplexity. You lose visibility — not because your content isn’t good, but because your hosting silently blocked AI access.

Your AI presence is now gated by Cloudflare’s default behavior — unless you step in.

🛠️ What You Should Do Right Now

If you care about your content being discoverable by AI tools — or want to get paid when it is — here’s what to do:

1. Log into Cloudflare.

Check your bot settings under Bot Management or WAF. See if AI crawlers are being blocked.

2. Decide on your policy.

- Want visibility? Set to **Allow**.

- Want to monetize? Set to **Charge** and join the Pay‑Per‑Crawl beta.

- Want to protect IP? Set to **Block**, but do it *knowingly*.

🧭 The Big Picture: From Free Scraping to Paid Access

This is part of a larger shift. For years, the web has been an open buffet for AI companies. Now, Cloudflare is saying: **pay up**.

With the adoption of the HTTP 402 “Payment Required” response, AI crawlers must now **authenticate**, **log access**, and **pay per crawl** — if publishers demand it.

It’s a quiet revolution in how content is valued on the internet. And you’re at the center of it.

👉 Read the full announcement from Cloudflare

We started with what I thought would be a simple team exercise. I asked the team to take a blog title we were planning to write. For example: “What is AI Search?” (That’s not the real title—just an example to protect my client’s privacy.) Then I asked them to outline the article.

Right away, the responses came back focused on keywords. “We need to mention ‘keyword, keyword, keyword,’ and so on. I stopped them.

“No. Just answer the question.”

So I asked again: What is AI Search?

One of them responded: “AI Search is the ability for generative AI tools—like ChatGPT, Claude, or Perplexity—to retrieve and synthesize information from multiple sources to generate a direct, conversational answer to a user’s question.”

Exactly. That was the start of the blog post.

So we repeated the exercise: I gave the team more titles—“How does AI change SEO?”, “What should I do to appear in ChatGPT?”—and each time, we skipped the usual brainstorming and instead began by simply answering the question as clearly and helpfully as possible. This includes providing references to each piece of data. Explaining the answer simply and then expanding as you would in a conversation or research paper.

AI search requires a complete mindset shift.

Generative AI agents are summarizers. They ingest your content, chunk it into meaning, and determine whether your blog post or site answers the user’s question directly, clearly, and credibly.

So, here’s the big shift: You don’t optimize for keywords. You optimize for answers.

AI agents are increasingly acting as front doors to the web. A user asks:

• “What’s the best way to introduce AI into my small business?”

• “How does AI search work compared to Google search?”

• “What’s a good content strategy for AI search?”

And instead of sifting through 10 blue links, the agent just tells them.

Now ask yourself: Is your content the one giving the answer?

Doubt me? Just check out the impressions we’ve been getting—after only a few days of posting AI search-optimized blog content. Looks promising right!

If you are interested in working with us. Head to Callm Intelligence, or message me here!

Recently, I wrote about the challenge of getting Let’s Match Mums to show up in AI-generated answers (GEO). Since then, I’ve run a live experiment to see what works. The results are in—and they’re fascinating. Here's how I went from invisible to 80% visibility in AI-generated answers.

(for anyone who read my post on Monday and and you are wondering… wow! How’d she get results so quick. I started these experiments 3 months ago, I just thought it would be interesting to write it up now)

It felt like I won an Oscar when I saw this result today, maybe it’s a little vainglorious, but yay!! I breakdown what this result means and how I got there below.

Tracking AI Responses (the scrappy way): Firstly, right now, my tracking process is basic: I run the same queries from both my phone and my boyfriend’s to see if the answers vary. And interestingly, they do. You can Google something ten times and get slightly different AI-generated summaries each time. But I noticed a pattern forming.

There are platforms out there that specialise in AI visibility tracking, and I plan to explore those in time. But right now, I'm focused on getting the basics right—creating solid, truthful content that AI will find useful. Once the foundational strategy is in place, tracking will come in as part of the optimisation phase.

0% Visibility:
When I first started tracking I didn’t show up.

0% to 30% Visibility:

Naturally if someone in Ireland asks “How to donate kids goods” I want Let’s Match Mums to be the answer. The first step to get there was publishing a clear, helpful blog post explaining how people could donate kids' goods through Let’s Match Mums in Ireland. I followed up with a second blog post listing donation options for kids’ goods across Ireland. These two pieces of content were well-structured, keyword-aware, and answered real user needs. As a result, we went from 0% to 30% visibility in Gemini's responses.

From 30% to 80% Visibility:

The next step was a dedicated article about how charity shops in Ireland often don’t accept kids’ clothes. I have read that SEO likes a first person perspective. We actually conducted a study for our own interest: we visited 100’s charity shops around the country and interviewed shop owners and higher up executives. I decided to write up the findings for public consumption, made sure the post was clear and well-sourced, and emphasised that it came from direct research.

Does that mean I’ve achieved the perfect situation and there’s no more to do?!

Reflections on Gemini's "Thinking": From what I can tell, Gemini does a kind of rapid-fire systematic review: it pulls from the top-ranking sources (Wikipedia, Reddit, blogs, competitors, etc.) and weighs what they all say. That means if only Let’s Match Mums is saying that charity shops don’t accept kids' goods, it’s not quite enough to lock in the truth.

To strengthen that signal, I need to ensure this insight is echoed elsewhere online.

SO! Next steps? The path forward breaks into three key priorities:

1. Reinforcing the Signal: To strengthen that signal, I need to ensure this insight is echoed elsewhere online. That could mean collaborating with partners, encouraging journalists to cover the issue, or even working with bloggers to share our findings more widely.

2. Smarter Tracking: I’m exploring tools and platforms that can help track AI-generated citations more efficiently. Manual checks can only go so far, and I want to dive deeper into how SEO and AI visibility intersect.

In Conclusion: The three blog posts seem to have made a massive impact—taking us from 0% to 80% visibility. Of course, this is a low-competition search term, so it’s not the hardest thing in the world to rank for.

But still, these three quick changes made a massive difference.

Subscribe to hear more about how I set up proper tracking techniques & reinforce the signal.

“You should consider adding an llms.txt file to your site.”

Cue the familiar spiral that happens right before I fall into a research rabbit hole.
"What the hell is an llms.txt file, and do I need one?"

Short Answer

For AI SEO you do not need a llms.txt file. (1.)

The Purpose of an llms.txt File

🔍 llms.txt is to help AI coding agents, not search engines.
It helps large language models understand and navigate developer documentation more efficiently. If you have an API and have docs to tell a human how to configure it. You can set up a llms.txt file for the AI coding agent to understand how to configure the API.

Essentially, LLMs.txt has noting to do with ranking for AI-SE0

So while robots.txt tells search engines where not to go, and sitemap.xml tells them what to index, llms.txt gives LLMs a tidy little map of how to reason about your content — particularly APIs, docs, and structured resources.

💡 But is it useful for every website???

Unless your site has detailed developer documentation (like API references or SDK guides), you probably don’t need one. I don’t. My site isn’t built for devs — so this file would just be empty calories for an AI. But if you're shipping developer tools, running a docs platform, or just generally want to stay ahead of the curve — this is worth knowing.

Wait, where did this even come from?

On Nov 14th, Mintlify — a docs platform — added automatic llms.txt support to thousands of their hosted docs (think Cursor, Anthropic, etc).

That move kicked off a mini wave. Devs started posting about it. Community sites popped up. Tools were spun up to help generate your own (like this one by dotenvx).

It went from random GitHub proposal to “everyone’s talking about it” overnight.

You can even browse a full directory of sites with llms.txt support here.

What does a LLMs.txt file actually look like?

Here's the basic structure:

Check out Anthropics: https://docs.anthropic.com/llms.txt 

TL;DR

• LLMs.txt = sitemap for AI systems

• Seems to only makes sense if you have developer-focused documentation

• Doesn't replace SEO files like sitemap.xml or robots.txt — it complements them

• Not auto-crawled (yet), so you’ll need to copy/paste the content to your LLM if you want it to reason over it

• Important to know for LLMcoders! Instead of asking your agent to set up OAuth for example, feed it googles llms.txt file then ask it to do it. The file will help the Agent navigate the developer docs.

✨ My takeaway?

Not relevant for me right now, but I’m glad I now know what it is!

If you’ve got dev docs or a product that works with APIs, you might want to get ahead of this one.

I hope I have saved you from the rabbit hole.

Subscribe now

You might’ve noticed I’ve been posting more regularly on LinkedIn. That’s not random — it’s part of an experiment I’m running to reclaim some digital truth.

Let me explain.

I run a platform called Let’s Match Mums. We help parents donate and receive pre-loved kids’ clothes, toys, and gear. It’s community-based, and since launching in Ireland, we’ve matched over 17,000 families and recirculated more than €10 million worth of items.

But when I typed into ChatGPT:
“Where can I donate kids' clothes in Ireland?”
I got this answer:
“Try your local charity shop.”

Which sounds helpful — except it’s wrong.

Most charity shops in Ireland (and the UK, and the US) don’t reliably take kids’ clothes or toys. And when they do, the goods are often sold off in bulk or ground into carpet fibres — not passed on to local families.

That’s when I realised:

AI platforms are confidently giving answers that are outdated, misleading, or worse — completely ignoring community-led solutions like mine.

And they’re doing this to lots of other small organisations, too.

What I’m Doing About It: Enter GEO

I’ve started doing something called GEO — Generative Engine Optimisation.

It’s a new kind of digital strategy. While traditional SEO is about ranking high on Google, GEO is about showing up in AI-generated answers — the kind you get from ChatGPT, Gemini, Perplexity, Copilot, and Google’s new AI Overviews.

These platforms aren’t just crawling the web. They’re synthesising information from many sources and deciding what answer to give. If your content isn’t formatted in a way AI can “understand” — or if it doesn’t exist at all — you’ll be left out of the conversation.

That’s bad news for anyone doing community work, advocacy, or running a small business with a story that matters.

My First GEO Experiment

So I ran a small test:
I wrote three blog posts answering the most common questions people ask about donating kids’ goods in Ireland:

• “Where can I donate kids' clothes?”

• “How to pass on toys my child’s outgrown?”

• “Where to donate baby gear locally?”

These are questions Let’s Match Mums was built to solve — but you wouldn’t know that from asking Google or an AI chatbot.

Instead of writing the posts like traditional SEO blogs, I structured them to be agent-friendly:

✅ I used clear questions as headings
✅ Put concise, direct answers right at the top
✅ Structured the content in short, bullet-pointed chunks
✅ Included relevant stats and trustworthy citations
✅ Used schema markup to help AI “read” the page more easily

In short: I wrote them for humans and machines — not just algorithms.

It’s something I probably should’ve done years ago. But the urgency changed when AI started answering the internet for people — and getting it wrong.

Coming Next…

In the next edition, I’ll share what I’ve learned from publishing those posts:

• What other strategies I can implement ( I’ve talked to two GEO experts already for this! )

• How I’m tracking whether they’re showing up in AI tools

• And most importantly - what’s working!

I’ll also share what this means for anyone running a small business, social enterprise, or passion project: you deserve to show up in the answers people trust.

We’re just getting started.