<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[Kevinleary.net]]></title><description><![CDATA[Kevinleary.net]]></description><link>https://kevinleary.hashnode.dev</link><generator>RSS for Node</generator><lastBuildDate>Wed, 18 Nov 2026 13:40:52 GMT</lastBuildDate><atom:link href="https://kevinleary.hashnode.dev/rss.xml" rel="self" type="application/rss+xml"/><language><![CDATA[en]]></language><ttl>60</ttl><item><title><![CDATA[Consolidating Agency Partners]]></title><description><![CDATA[Why working with one experienced consultant usually beats paying a handful of specialized agencies.]]></description><link>https://kevinleary.hashnode.dev/consolidating-agency-partners</link><guid isPermaLink="true">https://kevinleary.hashnode.dev/consolidating-agency-partners</guid><category><![CDATA[marketing]]></category><category><![CDATA[business]]></category><category><![CDATA[AI]]></category><dc:creator><![CDATA[Kevinleary.net]]></dc:creator><pubDate>Wed, 18 Nov 2026 00:20:07 GMT</pubDate><content:encoded><![CDATA[<p>More often than not, I see companies hire an SEO agency, a paid ads shop, a web development firm, an analytics vendor and a security or compliance firm.</p>
<p>On paper it looks like a specialized team. In practice it's five retainers, five account managers and a lot of finger pointing.</p>
<p>Most businesses don't need five agencies. They need one experienced person who knows how all of these channels and systems fit together, backed by AI agents that handle the repetitive work.</p>
<h2>The Big Agency Handoff</h2>
<p>During the sales process you meet the founders, the VPs and the senior strategists. Once the contract is signed, they move on to the next pitch and your account goes to a rotating team of junior staff juggling a dozen other clients.</p>
<p>You're paying a markup that covers office leases, middle management and sales commissions, and the people doing the work often have less experience than you do.</p>
<h2>Problems with Multiple Agencies</h2>
<p>Even when each agency is decent on its own, splitting the work means nobody owns the business outcome:</p>
<ul>
<li><strong>SEO agency:</strong> chases traffic numbers, whether or not that traffic converts.</li>
<li><strong>Ad agency:</strong> spends the budget on clicks, then blames the landing pages or tracking when leads don't close.</li>
<li><strong>Web dev shop:</strong> builds the site without much thought for technical SEO, page speed, analytics tags or conversion rates.</li>
<li><strong>Design agency:</strong> hands over a nice Figma file that doesn't help generate demand.</li>
</ul>
<p>You end up in the middle, sitting through weekly status calls trying to get four account managers to talk to each other.</p>
<h2>Agencies You Can Consolidate</h2>
<p>Here's what I typically replace:</p>
<table>
<thead>
<tr>
<th>Agency Type</th>
<th>What I Handle Directly</th>
</tr>
</thead>
<tbody><tr>
<td>Web Development &amp; CMS</td>
<td><a href="https://www.kevinleary.net/services/wordpress-consultant/">Custom WordPress</a> and VIP, <a href="https://www.kevinleary.net/services/headless-cms-developer/">headless CMS</a> builds with React and Next.js, <a href="https://www.kevinleary.net/services/laravel-developer/">Laravel</a>, <a href="https://www.kevinleary.net/services/webflow-developer/">Webflow</a>, Craft CMS and custom Stripe checkouts</td>
</tr>
<tr>
<td>SEO, AEO &amp; Content</td>
<td>Technical SEO, schema and content built for both search and <a href="https://www.kevinleary.net/services/aeo-ai-seo-consultant/">answer engine optimization</a>, so you show up in AI search results</td>
</tr>
<tr>
<td>Analytics &amp; Data</td>
<td><a href="https://www.kevinleary.net/services/google-tag-manager-consultant/">Google Tag Manager</a>, server-side tracking, Meta Conversions API, <a href="https://www.kevinleary.net/services/google-analytics-consultant/">GA4</a>, and pipelines into <a href="https://www.kevinleary.net/services/bigquery-consultant/">BigQuery</a> and Looker Studio</td>
</tr>
<tr>
<td>Paid Media, CRO &amp; Demand Gen</td>
<td>Lead generation, conversion rate optimization, ad platform conversion signals, and HubSpot or Salesforce integration</td>
</tr>
<tr>
<td>Security, Privacy &amp; Accessibility</td>
<td><a href="https://www.kevinleary.net/services/cloudflare-consultant/">Cloudflare</a> WAF and bot protection, Core Web Vitals, <a href="https://www.kevinleary.net/services/website-privacy-consultant/">GDPR and CCPA consent</a> with Consent Mode v2, and <a href="https://www.kevinleary.net/services/accessibility-consultant/">WCAG remediation</a></td>
</tr>
</tbody></table>
<h2>How AI Agents Fit In</h2>
<p>I've done this work myself for years, and even before AI one experienced person beat the multi-agency setup because the strategy stayed in one head and nothing got lost in translation.</p>
<p>AI agents make it a lot more effective. I don't mean typing prompts into ChatGPT for generic blog posts. I mean custom agents and automation workflows for audits, code scaffolding, vulnerability scanning, data cleanup and search data analysis. Work that used to take a junior agency team three weeks now takes minutes, and I spend my time on the parts that need experience.</p>
<h2>Experience Still Matters</h2>
<p>AI doesn't replace experience. Give it to an inexperienced junior staffer and they'll produce bad work faster. They won't know why a server-side GTM container is dropping events, why a WordPress query is slowing down checkout, or why an ad campaign is wasting money. Knowing that is the job, and the agents handle everything around it.</p>
<h2>Making the Case for One Person</h2>
<p>The hardest part of this isn't the work. It's making it clear that one person can be a highly effective resource across this many agency services, and often a better one than the agencies themselves.</p>
<p>Most people hear "I do WordPress, SEO, analytics, ads, security and accessibility" and assume jack of all trades, master of none. That's a fair instinct. Plenty of freelancers list everything and do none of it well.</p>
<p>The difference is that these services aren't separate skills to me. They're parts of the same system:</p>
<ul>
<li>Tracking is code, so it lives in the site build.</li>
<li>Technical SEO and AEO depend on how the site is built.</li>
<li>Consent gating decides what analytics and ad platforms can see.</li>
<li>Landing page speed and conversion rates decide how well ad spend performs.</li>
</ul>
<p>Agencies split these up because that's how they staff, not because the work is actually separate. The best way I've found to make the case is to start with one real problem, like broken attribution or a slow checkout, fix it, and let the results show how far the rest of it reaches.</p>
<h2>Benefits of One Consultant</h2>
<ul>
<li><strong>Lower cost:</strong> you pay for direct work, usually about half the monthly cost of an agency setup, without the overhead markup.</li>
<li><strong>No handoff:</strong> you work with the person doing the work, with no account manager in between.</li>
<li><strong>Everything connects:</strong> the ad, the landing page, the consent gating and the BigQuery reporting are built to work together from day one.</li>
<li><strong>Faster updates:</strong> changes that take three agencies a month of meetings and tickets ship in a few days.</li>
</ul>
<p>If you want to cut down your agency roster, take a look at my <a href="https://www.kevinleary.net/services/digital-growth-consultant/">digital growth consulting</a> or <a href="https://www.kevinleary.net/contact/">get in touch</a>.</p>
]]></content:encoded></item><item><title><![CDATA[Broken Attribution in Hubspot]]></title><description><![CDATA[If you're seeing HubSpot contacts with an original source of Offline Sources and no ad click history, an app sign-up is usually skipping the cookie.]]></description><link>https://kevinleary.hashnode.dev/broken-attribution-in-hubspot</link><guid isPermaLink="true">https://kevinleary.hashnode.dev/broken-attribution-in-hubspot</guid><category><![CDATA[HubSpot]]></category><category><![CDATA[analytics]]></category><category><![CDATA[marketing]]></category><category><![CDATA[attribution]]></category><dc:creator><![CDATA[Kevinleary.net]]></dc:creator><pubDate>Fri, 13 Nov 2026 13:18:12 GMT</pubDate><content:encoded><![CDATA[<p>If you're seeing a pile of HubSpot contacts with an original source of Offline Sources and no ad click history, the usual cause is an app sign-up that creates contacts through the API without passing along the visitor's HubSpot browser cookie.</p>
<p>I see this setup at a lot of SaaS companies. Marketing runs paid ads, HubSpot tracks the visits, and engineering wires the sign-up form into HubSpot. Everything looks connected, but the contacts that come out the other end have no ad click, no landing page, and no session history attached to them.</p>
<h2>A Common Example</h2>
<ol>
<li>Someone clicks a Google Ads ad.</li>
<li>They land on a landing page on <code>company.com</code>.</li>
<li>They click the call to action and go straight to a sign-up form.</li>
<li>The sign-up form lives in the app, usually on a separate domain like <code>app.company.com</code>, so it can't be a HubSpot form.</li>
<li>Engineering is asked to "connect HubSpot," so they create the contact with the HubSpot API using an upsert, or a search for the contact by email followed by an update or insert (same thing, just more roundabout).</li>
</ol>
<p>The contact gets created with whatever the sign-up form collected: email, name, maybe a company. What's missing is the link to everything that person did before they signed up.</p>
<h2>How HubSpot Tracking Works</h2>
<p>When someone visits a site with the HubSpot tracking code, HubSpot starts an anonymous profile for them, stored in the <code>hubspotutk</code> cookie. It records that they came from an ad, which pages they viewed, and what else they did. It just doesn't know who they are yet.</p>
<p>When that person submits a HubSpot form, HubSpot creates a contact and merges the anonymous history into it. If you open a contact record and look below the form submission in the timeline, you'll see every page view and ad click that happened before it.</p>
<p>A HubSpot form does this gluing for you. When you create contacts through the API you have to glue it together yourself. It's not difficult, but it's commonly missed.</p>
<h2>What Broken Attribution Looks Like</h2>
<ul>
<li>Hundreds or thousands of contacts with an original source of <strong>Offline Sources</strong></li>
<li>Contact records that only contain the fields from the sign-up form</li>
<li>Sometimes a little more data from billing or profile updates, because developers often sync any change to a user into HubSpot, but still no browser session data</li>
<li>Paid ad campaigns that look like they produce nothing in HubSpot reports</li>
</ul>
<p>The last one is the costly part. If you send conversions from HubSpot back to Google Ads or Meta for conversion optimized campaigns, those conversions depend on the ad click being attached to the contact. With no session data there's no click to attribute, so the ad platforms never learn which clicks turned into sign-ups and can't optimize toward them.</p>
<h2>Fix #1: Pass the HubSpot Cookie to the API</h2>
<p>This is the clean fix, and it's the way HubSpot documents it. Send the sign-up to HubSpot through the <a href="https://www.kevinleary.net/blog/using-hubspot-api-to-submit-forms/">Forms API</a> with the value of the visitor's <code>hubspotutk</code> cookie in the <code>context.hutk</code> field. HubSpot then treats it like a regular form submission and merges the anonymous session history into the contact.</p>
<p>One detail that trips people up: the CRM Contacts API (the upsert endpoint) has no field for <code>hutk</code>. Engineering can keep the upsert for syncing user data, but the sign-up itself needs to go through a form submission to carry the cookie.</p>
<pre><code class="language-javascript">// Server side, after a successful sign-up
const hutk = req.cookies.hubspotutk; // set by the HubSpot tracking code

await fetch(
  `https://api.hsforms.com/submissions/v3/integration/secure/submit/${PORTAL_ID}/${FORM_GUID}`,
  {
    method: 'POST',
    headers: {
      'Content-Type': 'application/json',
      Authorization: `Bearer ${HUBSPOT_TOKEN}`,
    },
    body: JSON.stringify({
      fields: [
        { name: 'email', value: user.email },
        { name: 'firstname', value: user.firstName },
        { name: 'lastname', value: user.lastName },
      ],
      context: {
        hutk,
        pageUri: 'https://app.company.com/signup',
        pageName: 'Sign Up',
      },
    }),
  }
);
</code></pre>
<p>A few things to check:</p>
<ul>
<li>The <code>hubspotutk</code> cookie is set on the root domain, so a server on <code>app.company.com</code> will receive it in the request as long as the HubSpot tracking code runs on <code>company.com</code>.</li>
<li>If the app is on a completely different domain, add it to HubSpot's tracking domains so the cookie carries across.</li>
<li>Create a form in HubSpot for the sign-up so the submission has somewhere to land. It's also a handy way to report on sign-ups.</li>
</ul>
<h2>Fix #2: Identify in the Browser</h2>
<p>If you can't get engineering time, or they won't change the sign-up flow, you can make the connection from the browser instead. After the person signs up, call HubSpot's <code>identify</code> method with their email on a page that loads the tracking code:</p>
<pre><code class="language-javascript">var _hsq = (window._hsq = window._hsq || []);
_hsq.push(['identify', { email: user.email }]);
_hsq.push(['trackPageView']);
</code></pre>
<p>The <code>trackPageView</code> call matters. <code>identify</code> only stores the email in the tracker and nothing is sent to HubSpot until the next page view or event. As long as the email matches the contact the server created, HubSpot ties the browser session to that contact.</p>
<p>It works, but it's a monkey patch.</p>
<h3>Where Identify Causes Problems</h3>
<p>Only call <code>identify</code> with the person's own email, at sign-up. Anywhere else in the app gets risky:</p>
<ul>
<li><strong>Invite a friend forms:</strong> the email typed in belongs to someone else, so identifying with it ties your user's browser to the wrong contact.</li>
<li><strong>Email changes in a profile:</strong> if you don't call <code>identify</code>, the contact keeps the old email, which breaks email marketing (one of the main reasons you're paying for HubSpot). If you do call it, HubSpot treats the new email as a new person and creates a brand new contact, even though the browser cookie is the same one already tied to the original contact. HubSpot's documentation doesn't mention this anywhere, but it's what happens in practice.</li>
</ul>
<p>Segment handles this with an <code>alias</code> call that links an old identity to a new one, which is one of the reasons I like it for this kind of work (more on that on my <a href="https://www.kevinleary.net/services/segment-consultant/">Segment consulting</a> page). HubSpot doesn't have an equivalent, so with this workaround it's best to leave profile updates alone.</p>
<h2>Fix the Root Cause When You Can</h2>
<p>I always push for fix #1. It's a small change for an engineering team, it follows HubSpot's documented approach, and it avoids every edge case above. The browser workaround is fine for getting attribution back quickly while you wait on engineering, but it shouldn't be the long-term setup.</p>
<p>If your HubSpot attribution is broken and you want help tracking down where the session is getting lost, take a look at my <a href="https://www.kevinleary.net/services/hubspot-cms-developer/">HubSpot development</a> work.</p>
]]></content:encoded></item><item><title><![CDATA[Bing Webmaster Tools for ChatGPT AEO]]></title><description><![CDATA[ChatGPT uses Bing search to provide citations, making Bing webmaster tools a significantly more valuable tool today.]]></description><link>https://kevinleary.hashnode.dev/bing-webmaster-tools-for-chatgpt-aeo</link><guid isPermaLink="true">https://kevinleary.hashnode.dev/bing-webmaster-tools-for-chatgpt-aeo</guid><category><![CDATA[SEO]]></category><dc:creator><![CDATA[Kevinleary.net]]></dc:creator><pubDate>Wed, 11 Nov 2026 02:24:24 GMT</pubDate><content:encoded><![CDATA[<p>ChatGPT uses Bing search to provide citations, making Bing webmaster tools a significantly more valuable tool today.</p>
<p>If you want to know why ChatGPT cites a competitor and not you, Google Search Console is the wrong place to look. I hadn't logged into Bing Webmaster Tools in years because Bing sends so little traffic. When I went back in, I found about 2,000 AI citations of my site in the last three months that I had no idea existed.</p>
<p><img src="https://www.kevinleary.net/wp-content/uploads/bing-webmaster-tools-ai-performance.png" alt="Bing Webmaster Tools AI Performance report showing citations and grounding queries" /></p>
<h2>Which Search Engine Each Chatbot Uses</h2>
<p>Chatbots don't crawl the whole web themselves. When one needs a current answer it runs a search in the background, reads the top results, and cites them. Each one leans on a different search engine:</p>
<table>
<thead>
<tr>
<th>Chatbot</th>
<th>Search Behind It</th>
<th>Where to Analyze It</th>
</tr>
</thead>
<tbody><tr>
<td>ChatGPT</td>
<td>Bing, plus OpenAI's own crawler</td>
<td>Bing Webmaster Tools</td>
</tr>
<tr>
<td>Microsoft Copilot</td>
<td>Bing</td>
<td>Bing Webmaster Tools</td>
</tr>
<tr>
<td>Gemini, AI Overviews, AI Mode</td>
<td>Google</td>
<td>Google Search Console</td>
</tr>
<tr>
<td>Claude</td>
<td>Brave Search</td>
<td>No webmaster tool, check rankings by hand</td>
</tr>
</tbody></table>
<p>OpenAI doesn't publish the exact mix, but the data is consistent. A Seer Interactive study found that 87% of ChatGPT search citations matched Bing's top results, compared to 56% for Google.</p>
<blockquote>
<p>If a page doesn't rank in Bing, it has a hard time getting cited by ChatGPT, no matter how well it ranks in Google.</p>
</blockquote>
<h2>What to Look At in Bing Webmaster Tools</h2>
<ul>
<li><strong>AI Performance:</strong> total citations, which pages get cited, and the grounding queries the AI ran in the background to find you. It covers Copilot and Microsoft's AI partners, which makes it the closest thing to ChatGPT citation data that exists.</li>
<li><strong>Grounding queries:</strong> these are not what people type. They are the short searches the AI writes for itself, and they are the terms to rank for in Bing.</li>
<li><strong>Search Performance:</strong> your Bing clicks, impressions and positions for those same terms.</li>
<li><strong>IndexNow:</strong> push new and updated URLs to Bing instantly. Cloudflare and most SEO plugins can do this for you.</li>
<li><strong>Site Scan &amp; URL Inspection:</strong> confirm the pages you want cited are actually indexed.</li>
</ul>
<p>One limit to know about: the AI Performance report shows citations, not clicks.</p>
<h2>Bing vs Google Ranking Factors</h2>
<p>Bing and Google share most of the same signals. The difference is how much weight each one gets, and Bing is the more literal of the two.</p>
<table>
<thead>
<tr>
<th>Factor</th>
<th>Bing</th>
<th>Google</th>
</tr>
</thead>
<tbody><tr>
<td>Exact match keywords in the title, H1, URL and meta description</td>
<td>More weight</td>
<td>Less, it matches on meaning</td>
</tr>
<tr>
<td>Social shares &amp; engagement</td>
<td>A stated ranking signal</td>
<td>Not a ranking signal</td>
</tr>
<tr>
<td>Clicks &amp; time on page from search results</td>
<td>A stated ranking signal</td>
<td>Downplayed publicly</td>
</tr>
<tr>
<td>Freshness</td>
<td>More weight, and IndexNow gets updates in right away</td>
<td>Depends on the query</td>
</tr>
<tr>
<td>Backlinks</td>
<td>Fewer needed, favors older and trusted domains (.edu, .gov)</td>
<td>Still the heavier signal</td>
</tr>
<tr>
<td>Named author &amp; cited sources</td>
<td>Listed under quality &amp; credibility</td>
<td>E-E-A-T</td>
</tr>
<tr>
<td>JavaScript rendered content</td>
<td>Less reliable</td>
<td>Handles it well</td>
</tr>
</tbody></table>
<h2>How to Rank Higher in Bing</h2>
<ol>
<li><strong>Use the exact phrase:</strong> put the query you want in the title tag, the H1 and the first paragraph, word for word. Synonyms that work in Google work less well in Bing.</li>
<li><strong>Write real meta descriptions:</strong> Bing reads them for relevance, not only for the snippet.</li>
<li><strong>Turn on IndexNow:</strong> and keep an accurate XML sitemap with real last modified dates.</li>
<li><strong>Update pages that matter:</strong> Bing rewards content that is current.</li>
<li><strong>Share on social:</strong> LinkedIn, X and Facebook activity counts for something in Bing.</li>
<li><strong>Serve content in the HTML:</strong> don't rely on JavaScript to render the text you want indexed.</li>
<li><strong>Claim Bing Places:</strong> if location matters to your business.</li>
</ol>
<h2>Use Both Tools</h2>
<p>I'm not replacing Search Console. Google still sends most of my traffic, and it's the right tool for Gemini and AI Overviews. But AEO work aimed at ChatGPT gets measured in Bing Webmaster Tools, and it's free. If you'd like help with this, I offer <a href="https://www.kevinleary.net/services/aeo-ai-seo-consultant/">AEO consulting</a>.</p>
]]></content:encoded></item><item><title><![CDATA[AI Powered WordPress]]></title><description><![CDATA[Make changes to your WordPress website in minutes instead of waiting weeks on an agency.]]></description><link>https://kevinleary.hashnode.dev/ai-powered-wordpress</link><guid isPermaLink="true">https://kevinleary.hashnode.dev/ai-powered-wordpress</guid><category><![CDATA[WordPress]]></category><category><![CDATA[AI]]></category><category><![CDATA[claude]]></category><dc:creator><![CDATA[Kevinleary.net]]></dc:creator><pubDate>Sun, 08 Nov 2026 20:18:10 GMT</pubDate><content:encoded><![CDATA[<p><em>Make changes to your WordPress website in minutes instead of waiting weeks on an agency.</em></p>
<p>Most marketing teams I work with have the same problem. Every new landing page, section redesign or headline change becomes a request to an agency, followed by a quote, a timeline and a few rounds of revisions. I set up WordPress sites so your marketing team can do that work themselves by asking Claude. You can see a working demo at <a href="https://bricks.kevinleary.net/">bricks.kevinleary.net</a>.</p>
<p><a href="https://bricks.kevinleary.net/"><img src="https://www.kevinleary.net/wp-content/uploads/ai-powered-wordpress-bricks-demo.jpg" alt="The bricks.kevinleary.net demo homepage" /></a></p>
<h2>How It Works</h2>
<p>I connect Claude directly to your WordPress staging site. Your team doesn't need Claude Code or any developer tools. They just work in the normal Claude app they already use, on a paid Claude plan. They describe what they want, and Claude builds it directly on your site. Because everything is built visually, your team can also click into any page in their browser to tweak text, swap images or adjust layouts by hand whenever they want.</p>
<h2>Visual Builders vs. Gutenberg Block Themes</h2>
<p>There are two ways to set this up, depending on where your site is today:</p>
<ul>
<li><strong>A visual page builder (like Bricks):</strong> this is the most flexible setup and what you see in the demo. If you're building a new site or ready for a redesign, a visual builder gives Claude full control over layout, styling and site-wide design changes.</li>
<li><strong>A Gutenberg block theme:</strong> if you already have a site built on WordPress's native block editor and don't want to rebuild it, I can connect Claude to it instead. I lock in your existing blocks, patterns and theme styles, and Claude assembles and updates pages using the block library you already have. You don't get the same deep layout control as a visual builder, but your team can still spin up landing pages and edit content using the blocks you already have.</li>
</ul>
<h2>Why Not Just Use Webflow?</h2>
<p>AI generated pages are a big part of Webflow's pitch today: describe a page and it builds it. This setup gives you the same thing on WordPress, plus everything that comes with WordPress:</p>
<ul>
<li><strong>No migration:</strong> keep your existing site, content, URLs and search rankings. Moving platforms is where a lot of SEO gets lost.</li>
<li><strong>Your choice of AI:</strong> use Claude, ChatGPT or whatever your team prefers, instead of the AI one platform decides to give you.</li>
<li><strong>You own it:</strong> host it anywhere, move it anytime, and never rebuild because a platform changed its pricing or plans.</li>
<li><strong>More room to grow:</strong> WordPress has plugins for almost anything, from ecommerce and memberships to forms, events and multilingual sites, without plan limits on how much content you can publish.</li>
<li><strong>More people who can help:</strong> WordPress runs a huge share of the web, so it's easy to find someone when you need them.</li>
</ul>
<p>If you like the idea of AI building your pages but don't want to give up control of your site, this is that. If Webflow is still the right fit for your team, I work there too as a <a href="https://www.kevinleary.net/services/webflow-developer/">Webflow developer</a>.</p>
<h2>Why the Design Framework Matters</h2>
<p>When a whole marketing team can build pages, there's a new risk: every page ends up looking like whoever made it. One person likes big bold headlines, another likes soft colors, and someone else adds a new button style for their campaign. Within a few months the site doesn't look like one brand anymore.</p>
<p>That's why the setup matters more than the AI. Before your team builds anything, I put a design framework in place:</p>
<ul>
<li><strong>A fixed brand palette:</strong> the colors your brand uses, and only those colors.</li>
<li><strong>Set typography:</strong> one set of fonts and sizes for headings, body text and buttons.</li>
<li><strong>Approved layouts:</strong> a library of sections and patterns your pages are built from, like heroes, feature grids, testimonials, FAQs and calls to action.</li>
<li><strong>Consistent spacing and buttons:</strong> so every page has the same rhythm and the same calls to action.</li>
</ul>
<p>The framework lives in two places. The first is the site itself, inside the page builder or block theme. The second is a shared Claude project your team works from, which holds your brand guidelines, voice and rules. That way every request starts with the same instructions, no matter who's asking.</p>
<p>Claude builds from that framework instead of from someone's personal taste. When someone asks for "a hero with a big headline and a register button," they get your brand's hero, not a new one invented for that page. Your team still has creative freedom in the copy, images and structure of each page, but the brand stays the same everywhere.</p>
<p>The framework also makes your brand easy to change. When your brand evolves, you update the framework once and every page follows.</p>
<h2>Checks That Protect Your Business</h2>
<p>Consistent design is only half of it. The bigger risk is content that misrepresents what you sell. AI writes confidently, and a confident sentence that's wrong about your product can cost you a customer, or worse.</p>
<p>Every business has different things to watch for:</p>
<ul>
<li><strong>Product accuracy:</strong> features, specs, pricing and availability that match what you actually offer.</li>
<li><strong>Claims:</strong> no promises, guarantees or results you can't back up.</li>
<li><strong>Language and style:</strong> your voice, your terminology, and words you never use.</li>
<li><strong>Compliance:</strong> required disclosures and disclaimers, and regulated language for industries like healthcare, finance, insurance and legal.</li>
<li><strong>Legal and brand risk:</strong> competitor mentions, trademarks and anything that needs legal review.</li>
</ul>
<p>I build these in as checks during setup, based on your industry and your rules. Claude reviews its own work against them before anything is published, flags anything that needs a person to look at it, and leaves anything sensitive as a draft for approval instead of publishing it. Your team moves fast, and the things that could get you in trouble still get a second look.</p>
<h2>What Your Team Does</h2>
<p>Type a request, paste a brief, share a screenshot of a layout you like, or say it out loud:</p>
<blockquote>
<p>Create a landing page for our spring webinar. Use our brand styles, put a full width hero at the top with the date and a register button, then three speaker cards, an agenda, and an FAQ at the bottom.</p>
</blockquote>
<ol>
<li><strong>Claude builds the page</strong> on your staging site, in your brand.</li>
<li><strong>Claude checks it</strong> against your accuracy, style and compliance rules.</li>
<li><strong>You review it</strong> in your browser on desktop and mobile.</li>
<li><strong>You refine it</strong> in the same conversation ("Make the hero bolder", "Tighten the headline", "Move the testimonials higher") or click right into the editor to make quick edits yourself.</li>
<li><strong>You push it live</strong> to your production site once everything is approved.</li>
</ol>
<p>Existing pages work the same way. Name the page and describe the change.</p>
<h2>Things You'd Normally Send to an Agency</h2>
<ul>
<li>Build a landing page for a campaign or event</li>
<li>Change the hero headline and image on the homepage</li>
<li>Add new FAQs to the support page</li>
<li>Duplicate a page for the next event</li>
<li>Redesign the header</li>
<li>Apply a new brand palette across the whole site</li>
<li>Swap the heading font everywhere</li>
<li>Pull the latest case studies into a filterable grid</li>
</ul>
<p>Site-wide changes take a single message too. Every page shares the same design framework, so the change reaches the whole site at once.</p>
<h2>Before &amp; After</h2>
<table>
<thead>
<tr>
<th>With an agency</th>
<th>With Claude</th>
</tr>
</thead>
<tbody><tr>
<td>Write a brief and send a request</td>
<td>Describe what you want</td>
</tr>
<tr>
<td>Wait for a quote and a timeline</td>
<td>See it built on staging in minutes</td>
</tr>
<tr>
<td>Send revisions and wait again</td>
<td>Refine it in the same conversation</td>
</tr>
<tr>
<td>Wait for them to launch it</td>
<td>Push it to your live site when you're happy</td>
</tr>
</tbody></table>
<h2>Security &amp; Guardrails</h2>
<ul>
<li><strong>Always build on a staging site first:</strong> I always recommend connecting Claude to a staging environment rather than your live site. Your team can experiment, build pages and test site-wide changes safely, then push approved work live, in one click on most managed WordPress hosts. I set up the push so it only moves pages and design, never your leads or orders.</li>
<li><strong>Access you control:</strong> each person's connection can be turned off at any time.</li>
<li><strong>Limited permissions:</strong> Claude can only do what you allow, and anything you don't want touched can be switched off.</li>
<li><strong>Easy to undo:</strong> page edits are saved as revisions, so you can always go back, and I set up backups before every push to the live site.</li>
</ul>
<h2>What to Work on With a Developer</h2>
<p>Handing day-to-day page work to your team doesn't mean the site runs itself. It frees up budget and time for the work that actually grows the business, and that work still needs someone who knows what they're doing.</p>
<ul>
<li><strong>AEO (answer engine optimization):</strong> this matters most right now. More and more buyers ask ChatGPT, Claude, Perplexity and Google's AI Overviews instead of clicking through search results. AEO is what gets your business named and cited in those answers. It means structuring your content, product details and expertise so AI tools understand them, trust them and recommend you. Most sites aren't ready for it, and the businesses that get there first will be hard to displace.</li>
<li><strong>Security at the DNS level:</strong> AI has made attacks faster and cheaper. Tools that once took a skilled attacker days now scan thousands of sites for weaknesses in minutes, and try passwords, forms and plugins with brutal efficiency. I always recommend Cloudflare in front of the site, which blocks bots, malicious traffic and attacks before they ever reach WordPress. It's the single best protection for a business site, and it makes the site faster too.</li>
<li><strong>Monitoring &amp; fixes:</strong> someone watching for errors, broken pages, failed forms, downtime and security issues, and fixing them before customers notice. A form that quietly stops sending leads can cost more than any agency invoice.</li>
<li><strong>SEO:</strong> technical health, site structure, page speed and search visibility. AEO builds on all of it.</li>
<li><strong>Conversion optimization:</strong> finding where visitors drop off, then testing pages, forms and calls to action so more of them become leads and customers.</li>
<li><strong>Analytics:</strong> tracking that's set up correctly, so you know which pages, campaigns and channels actually drive results, not just traffic.</li>
<li><strong>Ad optimization:</strong> landing pages, tracking and attribution that lower your cost per lead and show where your ad spend works.</li>
<li><strong>Integrations:</strong> connecting the site to your CRM, marketing automation, email and the rest of your stack, so leads land where your team works.</li>
</ul>
<p>Your team builds the pages. I make sure those pages are secure, stay working, get found, get cited, convert visitors and can be measured.</p>
<h2>Common Questions</h2>
<h3>Do I still need a developer?</h3>
<p>Not for day-to-day pages and edits. Your team handles new landing pages, copy changes and layout updates on their own. You still need a developer for the initial setup, and for the work that grows the site: AEO, security, tracking and integrations.</p>
<h3>Does it work with ChatGPT?</h3>
<p>Yes. Claude is what I recommend because it does the best job with design and following brand rules. The same setup also works with ChatGPT and other AI tools, so your team isn't locked into one.</p>
<h3>Can Claude break my site?</h3>
<p>Not your live site. Claude works on a staging copy first, and nothing reaches your live site until someone approves it and pushes it. It can only do what its permissions allow, and every page edit is saved as a revision you can roll back.</p>
<h3>Do I have to rebuild my site?</h3>
<p>No. If your site already uses WordPress's block editor, Claude can work with the blocks and styles you already have. A rebuild on a visual builder only makes sense if you want full design control, or you're already planning a redesign.</p>
<h3>What does my team need?</h3>
<p>A paid Claude plan and the Claude app, on desktop or in the browser. A Team or Enterprise plan lets everyone share the same project, so they all work from the same brand rules. No code and no developer tools.</p>
<h2>Get It Set Up</h2>
<p>If you want this on your WordPress site, take a look at the <a href="https://bricks.kevinleary.net/">demo</a> and <a href="https://www.kevinleary.net/contact/">get in touch</a>. I'll set up the staging site, the framework and the checks, walk your team through a real page from request to live, and stay on for the work that grows the site.</p>
]]></content:encoded></item><item><title><![CDATA[Real World MCP: Home Depot]]></title><description><![CDATA[How a Home Depot MCP server could give ChatGPT, Claude and Gemini live inventory, pricing and aisle locations, and turn AI chats into carts.]]></description><link>https://kevinleary.hashnode.dev/real-world-mcp-home-depot</link><guid isPermaLink="true">https://kevinleary.hashnode.dev/real-world-mcp-home-depot</guid><category><![CDATA[mcp]]></category><category><![CDATA[ecommerce]]></category><category><![CDATA[AI]]></category><dc:creator><![CDATA[Kevinleary.net]]></dc:creator><pubDate>Sat, 07 Nov 2026 15:28:37 GMT</pubDate><content:encoded><![CDATA[<p>Home Depot doesn’t have a public MCP server yet, so this is a look at how one could turn AI conversations into sales.</p>
<p>When a homeowner asks ChatGPT, Claude or Gemini how to fix a rotted deck board, the sale is often decided before they ever open a browser tab. Most AEO work so far has been passive: structuring web content so LLMs cite your brand. An MCP server goes a step further and lets the AI check stock, build a cart, and send the customer to checkout. I covered the basics of this in <a href="https://www.kevinleary.net/blog/mcp-discovery-servers-ai-agents/">MCP Discovery Servers for AI Agents</a>. Home Depot makes a good real world example.</p>
<h2>What a Home Depot MCP Server Provides</h2>
<p>MCP (Model Context Protocol) is an open standard that lets AI assistants query outside systems and take actions in real time. Instead of relying on old training data or web scraping, an LLM connected to a Home Depot MCP server could ask Home Depot directly for:</p>
<ul>
<li><strong>Local inventory</strong> at the customer’s preferred store or ZIP code</li>
<li><strong>Current pricing and promotions</strong></li>
<li><strong>Exact product names, SKUs and specs</strong> from the product catalog</li>
<li><strong>Aisle and bay locations</strong> inside the store</li>
<li><strong>Product compatibility</strong> (which dimmer works with which LED fixture, which trim matches which siding)</li>
</ul>
<p>Add account login through OAuth 2.1 and the assistant can act for the customer:</p>
<ul>
<li>Save a project list in their Home Depot account (or Pro Xtra account)</li>
<li>Add every part needed for the job to their cart</li>
<li>Give them a link for checkout, curbside pickup, delivery, or an in-store map</li>
</ul>
<h2>Example: Spotlight Bulb Replacement</h2>
<p>Say a homeowner uploads a photo of their backyard to Claude and asks:</p>
<blockquote>
<p>I’m replacing the lights on my backyard security spotlight and want to spread light more evenly across the yard. Here’s my current setup. If I want to replace just the bulbs, what should I get?</p>
</blockquote>
<p>Without an MCP server, the AI gives generic advice (“look for a PAR38 flood bulb with a wide beam angle”), and the customer probably buys whatever shows up first on Amazon.</p>
<p>With a Home Depot MCP server connected, it could go like this:</p>
<ol>
<li><strong>Photo check:</strong> the LLM looks at the photo and identifies the fixture type and socket base.</li>
<li><strong>Catalog lookup:</strong> it calls a tool like <code>search_compatible_products</code>, filtered to what’s in stock at the customer’s local store.</li>
<li><strong>Specific recommendation:</strong> it suggests an exact PAR38 LED bulb with a wider beam angle and more lumens, with current local pricing for the number of bulbs needed.</li>
<li><strong>Login:</strong> it offers to save the list or build the cart, and the customer logs in.</li>
<li><strong>Checkout:</strong> the customer gets a checkout link, or a list showing the aisle, bay and total price.</li>
</ol>
<h2>Example: Rotted Siding Repair</h2>
<blockquote>
<p>My siding is rotted in this spot (see image). What products do I need to replace this, and are they in stock at my local store?</p>
</blockquote>
<p>This one needs several products, not one: siding panels, J-channel trim, house wrap, corrosion resistant nails, exterior caulk, and maybe a pry bar.</p>
<p>When an LLM guesses at this list, the customer is left hunting down six items across different websites. An MCP server turns it into a single cart where every item is compatible, in stock at the same store, and ready for pickup. For a retailer, that’s where the real revenue is: complete project baskets, not single items.</p>
<h2>How LLMs Handle This Today</h2>
<p>Right now LLMs learn about retail catalogs through web crawls and live search. For a big retailer that leaves real gaps:</p>
<table>
<thead>
<tr>
<th>Capability</th>
<th>Web Crawls &amp; Search</th>
<th>MCP Server</th>
</tr>
</thead>
<tbody><tr>
<td>Inventory</td>
<td>Guessed or out of date</td>
<td>Live, store level stock</td>
</tr>
<tr>
<td>Pricing</td>
<td>Often weeks or months old</td>
<td>Current pricing and local promos</td>
</tr>
<tr>
<td>Products</td>
<td>Generic (“LED flood bulb”)</td>
<td>Exact SKUs, specs and compatibility</td>
</tr>
<tr>
<td>In-store location</td>
<td>None</td>
<td>Aisle and bay</td>
</tr>
<tr>
<td>Checkout</td>
<td>Customer leaves the chat to shop</td>
<td>Cart or list built in the chat</td>
</tr>
<tr>
<td>Competitors</td>
<td>AI may recommend another store</td>
<td>Your catalog is the answer</td>
</tr>
<tr>
<td>Attribution</td>
<td>Mostly invisible</td>
<td>Trackable tool calls and carts</td>
</tr>
</tbody></table>
<p>An LLM will never keep a live copy of a huge retail catalog on its own. Store stock changes constantly. If you want AI assistants to recommend your products with confidence, you have to hand them structured, accurate data directly.</p>
<h2>E-Commerce &amp; AEO</h2>
<p>An MCP server sits between two things that usually belong to different people:</p>
<ul>
<li><strong>For the CMO (AEO and e-commerce):</strong> as product research moves from search engines to AI assistants, your brand becomes the source the AI relies on to answer the question, and every cart it builds can be attributed. I wrote more about tracking this in <a href="https://www.kevinleary.net/blog/ai-traffic-attribution/">AI Attribution</a>.</li>
<li><strong>For the CTO (controlled data access):</strong> instead of fighting AI scraper bots hitting your servers, you offer a rate limited endpoint that exposes only what you want LLMs to see.</li>
</ul>
<p>Technically you’re setting up a separate system for people to buy your products, one that lives inside the AI tools they already use.</p>
<h2>Caching Keeps Costs Low</h2>
<p>The obvious question from engineering is whether millions of LLM tool calls would overload inventory and ERP systems. With the right setup the load is small:</p>
<ol>
<li><strong>Cache the static data heavily.</strong> Most of what the server returns (product descriptions, specs, compatibility, store layouts) rarely changes and can be served from a CDN or Redis cache.</li>
<li><strong>Cache inventory for short windows.</strong> Price and stock change often, but caching store level availability for 5 to 15 minutes absorbs big spikes without hurting accuracy.</li>
<li><strong>Separate read and write tools.</strong> Discovery tools (<code>search_catalog</code>, <code>check_store_inventory</code>) run without login at the edge. Cart tools (<code>create_cart</code>, <code>add_to_pro_list</code>) only run after OAuth login, so the expensive calls come from people ready to buy.</li>
<li><strong>Reuse the existing APIs.</strong> An MCP server doesn’t need a new backend. It’s a thin layer on top of the REST or GraphQL APIs that already power the mobile app and website.</li>
</ol>
<p>Compared to the sales it could drive, the cost to run it would be minor.</p>
<h2>MCP Server Setup Help</h2>
<p>If you’re the CTO or CMO at a company that could benefit from something like this, I may be able to help. Take a look at my <a href="https://www.kevinleary.net/services/aeo-ai-seo-consultant/">AEO consulting</a> work, or <a href="https://www.kevinleary.net/contact/">get in touch</a> to talk about how an MCP server could work for your catalog.</p>
]]></content:encoded></item></channel></rss>