Machine speed feedback loops

For years the first thing you would do when you were introduced to a new company or contact was google them or look them up on LinkedIn. That still happens, but people now also ask AI tools, so I wanted to specifically analyse how well potential clients could do that with myself and Patching. The feedback was ok, but there was material room for improvement, so I decided to dedicate a little time to it.

I added my entire website code repository to a Claude chat and asked Claude to review the entire codebase for SEO/AEO issues and inconsistencies, telling it to crawl the live site if necessary to confirm any findings. I work on the website in Replit on a cloned version of my master GitHub repo, so I told it to give me detailed technical prompts for any changes.

An afternoon's work

In about 5-10 minutes I had an extremely detailed analysis of the site and the improvements needed, along with 4 comprehensive technical prompts to execute the required changes. To give you an idea how comprehensive these prompts were, they were of the order of 150-200 lines per prompt, formatted as markdown but with code snippets and references to specific lines of code to change. Prompts 1, 2 & 3 were good to go, but I highlighted an issue with prompt 4 which was re-worked by Claude before I used it. Each one generated a text report in Replit that I could copy/paste back into Claude to iterate from should I want to.

After the iterations were executed and the site looked good in Replit, I pushed the code to GitHub and opened a Pull Request to merge into my main branch, which my separate admin account must approve, publishing the updated site live automatically within 60-90 seconds. Yes, this is still me controlling all these accounts and tools, but they provide quality gates to be sure nothing goes live without me double and triple checking it and limit the blast radius if my normal account is hacked. I have Patching's GitHub configured in such a way that it doesn't really matter if an LLM does something silly - my code will always be safe, and all code is security scanned before a PR merges.

Once live I checked with Grok, Gemini and ChatGPT to get their updated analysis, not bad!

• 9/10 for technical SEO from Gemini

• 9/10 for AI Retrievability from Gemini

• 8.5/10 for AEO (Answer Engine Optimisation) from Grok

Agency speed vs machine speed - own the code!

AI tools are good at reading and generating code, so the previous model of 'give me a CMS or an agency or something so that I can have a website without needing to know how to code' is completely flipped. Now you want to own the code and the hosting so that you can use whatever model is the flavour of the month to make speedy iterations on your site.

AI tools are great, but you still need someone to drive them, and that consumes time and requires enough domain knowledge to know how best to use the tools and make sure the output is high quality. The way to tackle that challenge depends entirely on a company's unique context, but it's probably no longer an agency with long cycle times - at the very least you want them to be able to make changes at machine speed too. You probably also want to seriously consider providing them the tools and having them work in your IT environment so that you still own all the code if you decide to switch them out.

Use multiple models

One thing worth flagging:

• When I gave ChatGPT the same prompt to check SEO and AEO that I had given Grok and Gemini, it confidently produced feedback based on a version of the site that was out of date. It hadn't fetched the live URL.

• I also didn't ask Claude for feedback as I have memory turned on and knew that could pollute the response.

 

The lesson there isn't 'ChatGPT is bad' or 'don't use Claude', it's that different tools and models have different characteristics, and they change frequently. AI feedback is only as fresh as the model's last web search or what’s in its training data, so it often makes sense to use multiple models and derive your answer from their combined responses.

 

Bottom line

Not every business owner, entrepreneur or even marketer is going to feel comfortable being this technically hands on, however the biggest change that has come with AI is that the 'make the computer do what I want it to do' magic is no longer restricted to software engineers, it is now available to everyone. As such getting your company systems setup so that your people can work at machine speed without breaking anything has become even more critical. Code management tools like GitHub are no longer purely developer tools, they are now core business tools, and understanding git commands and concepts is now a more broadly required business skill.

If you’re trying to make sure your team are well setup to take advantage of the benefits of AI, we can help. Reach out below.👇