SQ, a senior content marketer at Ahrefs, shows how he automated monthly refreshes of 12 data-driven blog posts using Ahrefs' AI agent Letter AI do, while stressing that AI still needs human guidance.
Key Takeaways
- Content freshness is increasingly important for both traditional SEO and AI search citations, and refreshes must be repeated continuously to maintain gains.
- Data-driven posts that rely on fresh data benefit most from automated, scheduled refreshes.
- AI automation can handle tedious refresh workflows, but humans must review outputs and guide the AI with clear, planned instructions.
- Starting with a small automation and iterating is a practical way to scale AI-assisted content work.
What the video covers
- Speaker SQ, a senior content marketer at Ahrefs, manages the newsletter and maintains roughly 150-200 blog posts, making manual content updates a full-time burden.
- Ahrefs' study of 17 million AI citations found AI assistants cite fresher content (avg. ~2.9 years old) than organic Google results (~3.9 years), though preferences vary: ChatGPT favors the freshest content, while Google AI Overviews favor older pages.
- Updating content produces traffic spikes (e.g., a link reclamation post rose from ~100 to ~1,000 monthly visits), but gains decay and must be repeated continuously.
- SQ's challenge was 12 data-driven 'showcase' posts (top Google searches, questions, expensive keywords, most cited domains) that need frequent fresh data to perform.
- Manual refreshes involve pulling data from multiple sources, filtering NSFW queries, formatting tables, creating images, republishing, and double-checking—tedious work that was done only quarterly.
- He automated the workflow with Letter AI do, which pulls and cleans data, shows removals, builds WordPress drafts, and emails him monthly for one-click review.
- Refreshing monthly instead of quarterly, the automation scales the refresh cadence to 144 updates a year without extra effort.
- AI still requires hand-holding: he recommends planning mode, clear instructions, iterative refinement, and starting with small automations before scaling.
- The approach has spread to the wider content team, with colleagues automating other content parts and reducing costs previously tied to design, programming, and data science work.
- Other tools mentioned include Codex and Claude Code as alternatives to Letter AI do.
Full Transcript — Download SRT & Markdown
Speaker A
Hello everyone. Hope you had a good night's sleep. Okay, thank you a later. [laughter] Yeah, so my name is Squiren or SQ for short. Everyone calls me SQ because it's a difficult name to pronounce.
Speaker A
Oops, I'm a senior content marketer at Ahrefs. I've been at Ahrefs for about seven years now. So my main responsibility is managing the newsletter and writing content for the blog.
Speaker A
So as you can see a big part of my work is actually to update all of the content I've written over the years, which is more than 150 close to 200.
Speaker A
So it is a big struggle because updating content is a full-time job on its own.
Speaker A
And we all know if you are working in SEO, content updating is really important.
Speaker A
And it's even true for AI search. So when we conducted a study on 17 million citations, we found that AI citations are actually fresher than organic Google Serps. So 25.7% fresher than organic Google Serps. And AI assistants actually cited pages
Speaker A
averaging only about 2.9 years old, whereas organic Google Serps are actually have pages averaging 3.9 years.
Speaker A
But it isn't uniform across all platforms. As you can see, ChatGPT prefers the freshest content, whereas Google AI Overviews and Google themselves prefer older content.
Speaker A
So ChatGPT likes very new content like 958 days since publication. And then it goes older and older after that.
Speaker A
And if you're an SEO, you already know this. Every time you update a piece of content, it works. So this is an example from one of our blog posts about link reclamation. So one of my colleagues actually updated the post in 2025
Speaker A
because the post was not updated for a long time. I think that post was updated in 2019 or 2020.
Speaker A
So when she updated it, you can see the massive spike right there. So it went from only about a hundred search visits a month straight to a thousand search visits a month at its peak in June 2025.
Speaker A
But of course, as you know, when AI search came in and Google also doesn't like non-fresh content. So by the time time went by, by mid 2026, which is this year, it has started to go down already.
Speaker A
So it has dropped by about 67% but still higher than before. So it was only getting a hundred visits a month.
Speaker A
Now it's getting 350 visits a month because we haven't refreshed it and also because of course everyone knows AI mode and AI overviews are eating all our clicks.
Speaker A
So you can clearly see updating content works, but the gain doesn't hold. You have to keep doing it over and over and over again. So updating works.
Speaker A
You just have to do it forever because every platform likes it. So it it only works if you can afford to keep doing it.
Speaker A
And if you're a content marketer like me, as you're publishing more and more and more content, it gets harder and harder and harder. If you're only publishing one post, updating it every month, easy. 150, okay? That's a that's
Speaker A
a whole year's worth of work. And then after you're done, you have to do it again next year.
Speaker A
So my challenge is to figure out a way to see if I can update this content automatically using AI.
Speaker A
So I found so at Ahrefs, because we are a data big data company, we want to promote our data. So we actually have a whole bunch of blog posts kind of like selling our data, showing off our data.
Speaker A
So these are blog posts like top Google searches where we show you the top Google search across the US and all over the world. Top Google questions, the same thing but questions only. Most expensive keywords, which are all the
Speaker A
keywords in our database is most expensive, the highly, highest bidded on in Google Ads.
Speaker A
Most searched people is a entity based post where we take the top searches and we filter for all the entities for people and so on and so forth. And then we have the most cited domains post for every platform that we have. And all
Speaker A
together this is like 12 posts. This is just the tip. Okay, this is just what I do. My colleagues have their own version of posts as well. And as the time goes by, we have more of this. So this is a
Speaker A
challenge. This kind of post is only useful if the data is fresh. But I can't spend my whole work hours refreshing this content all the time. So this is a good candidate for automating and that's I thought.
Speaker A
So updating sounds like easy, right? Whenever you read a blog post, update your content, update your content every month. It it sounds like a senior guy just telling you, you know what, just control C, control V something and put
Speaker A
it inside. But real updating is difficult, right? You need to figure out what went wrong. Why is your traffic going down? What do you need to change? Is it new images, more data, new content? Do you need to add new
Speaker A
subheadings, so on and so forth? So actually real updating work is very difficult. Or not very difficult, is very tedious. Right, so for those blog posts, for those 12 blog posts, I have to pull fresh data from multiple locations because we had them
Speaker A
all in different places. And then I had to export them. I to make sure there's no junk. I to make sure there's no NSFW queries because we do not want to show them on our content.
Speaker A
And then I have to format them as a table. I to make new images and wait for design to make the new images for me. And I have to make sure everything looks okay. I didn't accidentally destroy anything. And then
Speaker A
I have to republish it, change the date, publish it again. And then double check everything is okay again. And then next post and then next post.
Speaker A
A lot of the work is actually really really boring, nothing exciting for a content marketer of many years.
Speaker A
The only thing you need to do is make sure everything is going well. Right. So the more you publish, the more it's a slog, right? So I can tell Ryan, my manager, I just updated a whole bunch of posts. He can look at it and he says,
Speaker A
has anything changed? And to be honest, not much, but I did my work. So it's a little bit hard to justify. So it will be better if it's done automatically.
Speaker A
So too much work, we did it every quarter, even though actually we wanted to do it every month if possible.
Speaker A
Right. So this is another example, top Google searches. Every time we updated it, it spiked, as you can see. So Google actually, for this post, Google loves it to be the freshest as it can be. Every time I wait for a
Speaker A
three months or six months to republish it, it drops like crazy. And the moment I republish it, boom, it goes up.
Speaker A
So this has to be the freshest possible. So that's what I did. I went to Letter AI do, our new AI marketing agent from Ahrefs. So Letter AI do has every single Ahrefs data point built inside.
Speaker A
So I can ask it to do anything. So I tried to ask it to do my automation for me, which fortunately succeeded.
Speaker A
Right. So what it does for me right now, Letter AI do, it pulls all the data from every the correct source. It cleans it for me and shows me shows me what it removes. It then even built because Letter AI do is
Speaker A
we can connect it to WordPress. So it even builds the entire draft for me in WordPress. And all I have to do, oh sorry, and then it emails me at the first of every month to tell me the post
Speaker A
is ready for review. And then I just need to click those posts, okay, and it's done.
Speaker A
Oh by the way, you can do this. I'm just doing it Letter AI do, but if you have Codex or Claude code or whichever thing you do, you can do it as well. It just I prefer Letter AI do
Speaker A
because of course I work for Ahrefs and secondly, it has all the data that I need.
Speaker A
Right. So this is the result. Now I can publish it every month uh for the scheduled refreshers a year is now 144.
Speaker A
I can do it easily. I just need to take a quick look, press okay, and that is it.
Speaker A
Right. So I thought it was a just a chore that I automated, you know, like sounds very boring. Wow. So I but I anyway I told my team and now they are all inspired and actually the whole content marketing
Speaker A
team is automating different parts of their content right now. So my colleague Louise has also a bunch of similar blog posts where she's talking about the fastest growing companies, fastest growing SaaS companies, a little bit of a programmatic SEO play, but still very
Speaker A
heavily content-based, so a little bit tedious and difficult to update. But right now it's done automatically by let AI do.
Speaker A
Right. So uh I'm going to show you like what I did basically. Okay. So one of my fears when AI came was that I didn't know what to automate. I think that I think there's a lot of AI panic.
Speaker A
Uh a lot of managers are telling their team use AI for something, automate something, and then they just leave them to be. And then everyone has AI panic.
Speaker A
Okay, I'm supposed to do something. I don't know what to do. I don't know what to do. What should I be doing? What should I be automating? I've been trying this and trying that and everything is failing.
Speaker A
And because that is the truth, automation sounds really nice, but actually there's still a lot of work to put inside to make AI do the thing you want to do. Secondly, even if uh it automates to this extent, there is
Speaker A
still a lot of review to be done. So your work just changes. Still a lot of time spent having the review.
Speaker A
So you want to actually start by automating something you already understand. You are already doing it for a very long time. It's just been really tedious. You really hate it. You think it's not worth your pay grade. You just
Speaker A
want to do something better. So for me that was uh those blog posts that I I mentioned.
Speaker A
Uh it is interesting to readers if data is fresh, but it's not interesting to me because basically I'm doing really, really boring work, just copy pasting data.
Speaker A
So it was a good candidate for me. It is a candidate that runs repeatedly.
Speaker A
I have to do it all the time. It's really stable. It doesn't really change much the the blog post itself.
Speaker A
Where there's clear value because it has a lot of search traffic and it sells our data. So it has a lot of business potential to keep this post updated.
Speaker A
And because I've done it for so many years, I already know what is good and what is bad.
Speaker A
Okay so everyone knows white coding today. So you don't actually need to know how to code anything, any tool, anything. You just need to know you just need to have good judgment, good taste, know what your work entails, what needs to be
Speaker A
done, and what is wrong. So I started with the chore, my challenge, 12 blog posts, boring updates.
Speaker A
And then before you do anything, you want to know what is good because sometimes a lot of times when we we are so used to doing the thing over and over again, we already know what is good, but
Speaker A
we have never tried to write down or tell somebody what good means. So you want to well ideally you don't want to go straight to AI and tell and start writing things out. What actually yesterday I just read a blog
Speaker A
post by somebody from Animals, which is a famous US content marketing agency. He has even gone back to using pen and paper to first write down what he wants before he goes straight to AI because when you start conversing with AI, he
Speaker A
has the ability to make you stop thinking and stop direct you kind of stop directing it in the direction you wanted in the first place because you would never clear what you wanted. So you wanted to sit you
Speaker A
want to sit down and go somewhere offline matters, pen and paper. Uh what I do today is that I actually like to I use Whisper Flow, which is a AI transcription service.
Speaker A
I actually just kind of put on my headphones, close my eyes, and just start talking for like 20 minutes into AI just to kind of like try to figure out what I want to say. I don't even type
Speaker A
anymore today. I mostly talk to AI, which is very weird when I go to a co-working space.
Speaker A
Yeah I I know I'm just talking okay. If you see me somewhere, ignore me.
Speaker A
Yeah, so I talk I tell you what I do. I try to remember what the entire process is. Oh, you need to take the data from this place, you need to take the data from that place, you need to do this,
Speaker A
you need to clean this, remove that, blah blah blah. Okay so and this is what happens. I defined good. I said okay, you're not supposed to include this in this table, you could exclude that in that table, so on and so forth.
Speaker A
Right, so this is what I do. Okay, so this was my process in the first place, monthly trigger, that's why I chose the first of every month to receive the email for review, the input which is all the keywords and
Speaker A
all the data that I needed. Actions, what do I want the AI to do?
Speaker A
Okay, I want the AI to extract the data, I want the AI to clean it for me, I want the AI to remove junk. Ideally, I want the AI to automatically identify what I do not want. So I have to
Speaker A
continue to teach it what is good and what is bad. And then that comes from checking the output over and over again.
Speaker A
I'll show you an example later what went wrong. And then of course, you have to have judgment as a person who is an expert and then you hand off.
Speaker A
I never let AI publish it for me even though ideally I want to because I I'm still a little bit afraid of what can go wrong.
Speaker A
But actually with lay I do you can you can I can directly ask lay I do to just publish without my interference, but still I want my job. So [laughter] I hope that it's not going to screw up the whole
Speaker A
block and then tomorrow you'll see that I'm from somewhere else. Okay, so this is the brief that I gave it.
Speaker A
What I just repeated, the inputs, what it should look like, what it should never change, and examples of what is good and bad.
Speaker A
Okay so I never go to uh executing because uh I find that sometimes I waste too much time trying to asking to change a lot of things.
Speaker A
So I actually prefer to ask AI to plan first. So I will tell you what I want and then I will ask it to show me the plan. There is a plan mode in most tools. I think Cloud Code there's a plan
Speaker A
mode, Codex there's a plan mode, Later I do there's a plan mode or brainstorm mode. So I want to see what it's trying to do first before it does it. Otherwise I mean I have spent like countless hours
Speaker A
scold scolding AI. I hope that robots don't go after me but I have spent countless hours scolding AI. No, this is not what I want. Please do it again.
Speaker A
Yeah, so plan first. Uh do one stage first. Don't jump straight in and automate every single thing. You want to make sure something is working to your liking.
Speaker A
Then you make the process work for other things. And another thing that I learned actually from Ryan who actually made a content pipeline. So he actually made a content pipeline that could take a keyword and draft the whole post right
Speaker A
away to publishing. So uh but what he did varies in a very smart way is that he actually made sure there was an output at every stage because I think a lot of people when they try to automate something with AI,
Speaker A
they just want to see the final output which is uh which is okay given the AI is is continually improving.
Speaker A
But sometimes along the way AI just kind of either it compacts the context too much or it forgets its context or it just kind of went off the rails and do something different.
Speaker A
So what you want to do is that you want it to uh create at every stage that you are doing something you want to have it create an output so that you can check that at every stage it is doing the correct
Speaker A
thing. And every time something is wrong in the final result, you can actually go back to each output to see where it went wrong. Then you can just make a minor change rather than having to always restart the process
Speaker A
and start a new chat, do it again, you know, so on and so forth.
Speaker A
So for this particular process that I was doing, uh it always shows me the raw data somewhere in the tool that it built.
Speaker A
always shows me what it removes. It shows me uh what it classified, like example for most searched people what kind of entities it's reading.
Speaker A
Uh and then before it even uploads to WordPress, it shows me the formatted table and then it shows me the draft. So anytime anything goes wrong, I can check each stage to see what went wrong.
Speaker A
Okay, so it sounds really good, but I have to be honest, you have to hand hold AI still. We are not at a stage where AI can do everything for you.
Speaker A
So for example, uh when I tried to automate top Google questions, it gave me a list of questions. Amazing. I thought I was done. And then after reviewing, it started including a lot of uh movie titles like How to Train Your
Speaker A
Dragon, uh when Life Gives You Tangerines, the famous K drama, because it has question words. How to train your dragon?
Speaker A
When Life Gives You Tangerines? So AI doesn't automatically know that these are not questions. So this is where your judgment comes in. You have to keep reviewing the output and keep telling AI, okay, this is not what I want. Uh you
Speaker A
are kind of on the right track, but you know, this is uh not a question. Change it.
Speaker A
And then more things like because I I wanted top Google questions, so AI also automatically thought anything with the word questions is a question. So again, you had to tell it, okay, interview questions are not technically a question. It is ques
Speaker A
it is a question about questions. So it's not technically a question, so please remove it from the list. So again, this is your judgment. You know the work. You know what is good, what is bad, what is the correct answer, what do
Speaker A
you want? That's why I say before you do anything, you really need to be clear what you want because you're not going to jump into AI and ask it to do a lot of weird things. I mean, you can for fun, but not for work.
Speaker A
Yeah. Okay so it is really fun to set for example Cloud code on auto mode and let it do its thing.
Speaker A
Uh but again, I like my job. So I let AI do most of the boring work. Okay, so it does all of the boring work and all I do is to sit there like a like a manager using my eyes to tell it what
Speaker A
to do and what is wrong and what is correct. Uh my dream. No, I'm just kidding.
Speaker A
Yep. So again, what I said, make sure you have a pilot first. I To be honest, these are not new ideas.
Speaker A
These are new ideas to us, but these are not new ideas in general to humanity because anytime somebody commissions for example a TV show, they always make a pilot first. No, is very rare that for example a network or even Netflix will
Speaker A
jump straight to making a drama that is a 20 episodes long without knowing whether anyone would like it. So these are actually really really old ideas that we can reuse for uh even in a modern era of AI, do something uh try
Speaker A
something first, make something small. If it works, upgrade it. So that was my personal challenge. I just want to show you like multiple examples of what you can potentially automate.
Speaker A
So for SEO, monthly performance report. So uh personally at Ahrefs, my manager has to report to Tim, our CMO every month on the blog.
Speaker A
So obviously reporting is one of the most boring things anyone can do, even though it is very important.
Speaker A
So what uh Ryan has done is that he basically uh went to let I do again and gave it GSC data, Ahrefs data and everything and now it generates a report that automatically sends it straight to Slack so that uh
Speaker A
Tim can see it. Uh weekly site audit, if you use AI do, you don't have to because we have MCP as well, so you can connect Hrefs MCP to Cloud code or Codex. But with let I do uh Uh one of our developers did was that
Speaker A
he actually took uh he actually set up a automated crawl every week of the Ahrefs website.
Speaker A
So what it does is that the site audit now actually automatically detects issues, creates a GitHub PR for any uh developer to immediately fix it.
Speaker A
So that's what we have been doing personally ourselves. We have been uh eating our dog food so to say.
Speaker A
So we have made a lot of different things. We have uh a content we have a content decline tool that we can see which posts are actually quietly decaying in search traffic so that we can update it automatically.
Speaker A
Uh you can find like if you are still building links, you can actually ask Later I do to show you every week all the high DR backlinks that you have lost.
Speaker A
And also Reddit is very important and we have a lot of Reddit data in Brand Raider and Site Explorer as well. So you can easily just ask it for all every week ask ask it to give you a list of
Speaker A
Reddit threads that actually started ranking for the keywords or prompts that you care about.
Speaker A
Content yeah similarly content decline. We also use a Later I do to ask uh what went wrong and uh what we need to change and can you automatically write the excerpt or the specific paragraph that we need to add
Speaker A
in order to rank better. Uh this is Ryan again. He every morning he can look at his tool and see all of our competitors latest posts and even even has a I like it and I dislike it kind uh check mark beside each post so
Speaker A
that he can just when he clicks I like it it will just automatically save inside the tool as content idea for our own blog.
Speaker A
And so on and so on. Internal links you can ask LLMs to automatically tell you which links you should include in every new post using some scientific uh thing you can you will look at the entities beside the
Speaker A
phrasing and everything. It's up to your imagination basically. AI search use Brand Raider data to show you where you are showing up, which citation you need, are you being recommended, are you being mentioned, so on and so forth.
Speaker A
Yep, so the method again. Okay. Yeah, so I just want to say this is a idea that we've been thinking a lot about at H Reiss, which is that I think when everyone is using AI, their immediate thought is
Speaker A
cost savings. I want to use AI to do more with less money. I mean, it's fair, but cost I I think at H Reiss, how we think about it is that cost is a very I don't want to use the word low level,
Speaker A
but it's a very basic way of understanding how to use AI. Because ultimately, as we all know right now, if you go to Google or ChatGPT, for a very long time, they were even citing a lot of slop. So
Speaker A
basically, there was a snake eating its own tail process where ChatGPT is basically citing itself because it's citing the slop made with a ChatGPT and then feeding back itself and then doing it over and over again, which is why in I think in the past two weeks,
Speaker A
people have noticed that ChatGPT has some kind of newer measures when it comes to RAG, Retrieval-Augmented Generation, is doing a lot more site searches. So basically, it what people are suspecting is that ChatGPT the engineers themselves has given
Speaker A
ChatGPT, the tool, maybe a short list of good websites that it should automatically consult.
Speaker A
So it's just going straight to doing those site searches rather than running a full query fan out to find different things.
Speaker A
So you you don't want to use AI as a cost measure to make more content that is faster, that is worse, and then just because you can show your boss or your team that, you know, you are publishing
Speaker A
more. Maybe last month you were publishing five, this month you published 100, next month you published 10,000, which is not good for anyone in this world. As consumers, we hate slop.
Speaker A
Uh I mean only businesses like slop basically because it looks like they're doing something. Everyone hates slop, so you don't want to do more slop. I think it's a terrible way of using AI. So what we are doing right now is that
Speaker A
we have used AI to save a lot of time. We want to use AI now to make something that we previously couldn't do.
Speaker A
So in the past they probably we had a lot of content ideas that we needed design work, a programming work, a data science work that we can now as a content marketer do it ourselves with Codex, with Cloud Code, with Let AI do.
Speaker A
So now we want to use those tools to make something that is better, that looks more effortful, more beautiful, better UX, better UI, maybe embed a quiz in one of the posts. So in the past if you wanted to have a quiz in a in a a
Speaker A
blog post, you had to beg, I mean don't have to beg, but you have to put in like PR for your developers and then they will forget it two years later and then you're like, oh my god, what happened to
Speaker A
this? Oh, we were too busy and then you never get anything done. But now with Let AI do, we can easily build quizzes.
Speaker A
We can build free tools, which is what we did recently. We built a LLMs.txt free tool by ourselves without any developer work.
Speaker A
We can now include custom images that could explain what we wanted to say in a blog post much better than before. We don't have to wait for design resources.
Speaker A
We can maybe even make comics, make books to make our content look better. So I think this is at the end I'm teaching you how to automate your work, but I also want you to use AI to do better work, not do more
Speaker A
work at lower cost. Okay, so just these are resources if you're interested. I have a booth outside as well, you can come and talk to me.
Speaker A
So the first one is the tool itself. The next two is just the various examples that I showed of how to use Let AI do.
Speaker A
And thank you. Yeah, this is all. [applause] Thank you, SQ. I think we have time for a couple of questions. So does anyone have any questions for SQ?
Speaker A
No questions. Sorry? When do you prioritize a brand new page versus a update? Honestly speaking, we don't really have a like a like a I don't know, like machine like a SOP for that. It's kind of like 50/50. We
Speaker A
try to update as much old posts as possible, and then we also try to write new posts. We have some kind of KPI for that.
Speaker A
I mean, sorry, it's a terrible answer to your question, but we we don't really have that kind of prioritization address.
Speaker A
Yeah. If you really want to see prioritization, we always look at business potential. There is a internal rubric that we use.
Speaker A
So business potential is invented by Tim. So it's just a very kind of basic way of scoring each topic's potential. So zero to three.
Speaker A
At three is about like this topic can promote our any of our feature of our product or our data.
Speaker A
And then it goes down. Zero is where you can pitch anything at all. So we try to always do work that scores a three first, brief, update, or new content.
Speaker A
Does that make does that make sense? Yeah, I like that score concept a lot.
Speaker A
Yeah, so we we always try to score along those lines. So even up like recently we actually had a meeting that said we need to update a lot of our older content because AI search is moving so fast that
Speaker A
a lot of our old pieces on GEO and AEO and AI search are already old six months later.
Speaker A
So we already actually went back and said saw that these are all business potential three blog posts for our Brand Raider tool, right? So we are going back to update all of them again as part of our current KPI.
Speaker A
Yeah. Any more questions? No? Then we'll wrap it here. Thank you so much SQ.
Speaker A
[applause]
Topics:SEOcontent marketingcontent automationAI searchcontent refreshAhrefsAI agentsAI citationsworkflow automationWordPress











