This AI Tool Does Keyword Research For You

[00:00] What if you could type just one word and have all  of your keyword research done for you? And I mean   everything. You can see the keywords that you  should go for. You can see them organized into   clusters. You can generate content briefs at  the click of a button. And while those briefs  

[00:16] are generating, visually see exactly how these  clusters should be internally linked throughout   your site. There's no tool that does this well  and fast until about 2 days ago. I built an AI  

[00:28] tool that can do all of this, hand you the entire  keyword road map for any niche, hundreds of hours   of work packed into about 20 minutes. And today,  I'm going to show you how to build it yourself.  

[00:40] All you have to do is just type in one word. So,  I'll do house plants and run the research. And   right now, it's looking for seed keywords. It's  going to go in and expand those keywords. And now,   it's filtering the keywords. And this is where  the bulk of the time goes because AI is actually  

[00:56] processing every single keyword as it goes through  to make sure that they're relevant first and then   it's running different at calls right here. It's  at 19% and it's going to keep going to make sure   that we're getting the exact keywords that would  be relevant to in this case the niche of house  

[01:12] plants. And after this happens, the clusters are  going to form. It's going to generate an executive   summary and tell you exactly what happened. So  while this is happening, let me show you the   prompt that I used to build this. So I started off  this chat session by saying, "Build an automated  

[01:27] keyword research tool using Opus 4.7. The user  inputs a niche. The tool generates seed keywords,   clusters them into groups, and produces a  comprehensive report with an action plan. In terms   of output, all I wanted was keyword clusters with  search intent classification. Super important. A  

[01:43] competitor gap analysis and a visual map showing  internal linking structures between clusters.   And this part is usually the part that takes a  really long time. So I said I also said that I'm  

[01:56] open to additional suggestions during planning and  that's the only prompt that I used. Now this tool   didn't come from one hero prompt. I spent days  fine-tuning this and over $200 in API tokens,  

[02:09] which is really small in the grand scheme of  things considering what it's able to do. Uh,   so rather than just reading my entire agent  a chat to you as a really bad bedtime story,   I'm going to highlight some of the biggest  problems and I'll explain exactly how I fix them  

[02:25] so you can build it, too. If you run into the same  problems, some of them you might, you can build it   with the exact same specs that I did here. Now,  the way that I built this tool is with agent A.   And if you don't know what Agent A is, it's an AI  agent with unrestricted access to HF's data. So,  

[02:40] it can actually pull real SEO, Google, and  marketing data, which is obviously super   powerful for any marketer. And you'll see why  that the unrestricted access part matters so much,   especially for things like SER data as we go  through the build. So, it starts by asking me  

[02:56] where I want this tool to live. And so, I wanted  it as a console app. And so you can see that I   built it here. Um, and we can access this. I can  access the keyword research tool just like that.   And then it asks where we want to source the data.  So I said keywords explorer as well as HRF site  

[03:11] explorer. So we can actually find uh keywords that  aren't necessarily so obvious but competitors are   ranking for. Then it asked me about the competitor  gap analysis like how should the competitors be   picked? We're basically just taking data from  the organic competitors report in hrefs and  

[03:25] how big should it be? Uh, not that important. um  beyond what you listed which of these are useful.   So it kind of gave me this checklist of different  things and I told it the prioritization score per  

[03:37] cluster is important and the way it's scoring it  is based on volume intent and difficulty. Quick   wins is always good because especially when you're  starting a new site, you don't want to compete for   things that are out of your league. And so getting  a few quick wins is always a good thing to do. So  

[03:52] uh low difficulty is one that I flagged. And then  I want to be able to export this as PDF and CSV.   What style for the cluster link map? And I chose  the hub and spoke diagram. And this is something  

[04:06] that was popularized by HubSpot many years ago  where we have, you know, a pillar content in the   center and then supporting pages around it. So  that's basically this one. Yeah. So this is the  

[04:18] whole idea of clustering. So, we have something  like this where we have a pillar page about   putters and then we have spokes that come from it.  So, you know, best putter, hockey stick putter,  

[04:30] Tiger Woods putter, Scotty Sheffler putter, some  different ideas that we can go after there. All   right. And then it takes the plan, it goes,  it tells you everything that it's doing. It's   basically like somebody who's thinking out loud.  And it'll go, it'll build the UI. It has more  

[04:45] questions as it's going because it's processing as  it goes. I want us data. I want uh competitors to   be from the top five and then we're going to cue  it and notify when ready. Then it goes it builds  

[04:58] me the first version of the tool and then that's  when we run into real problems. So before we get   into the problems I need to explain the keyword  research process to you because everyone does it   slightly differently but there are certain pillars  that need to be done. So the first step is finding  

[05:13] seed keywords. So for a niche like golf that  might be something like golf, putter, irons,   driver and we can run a search. So this is what  happens. Now from this search we need to go and  

[05:26] generate a whole bunch of keywords. So a keyword  research tool will give you a whole bunch. But the   thing is these are not going to be the queries  that you're actually going for. So you need to   actually filter these down. There's 4 million  keywords in here. That's crazy, right? So you  

[05:39] might do that by looking at questions related to  these queries. So, like what is a golf handicap?   How do I That's irrelevant because of driver,  a golf driver versus Door Dash driver. We'll   questions or you might actually use modifiers. So,  a modifier is just an add-on to a base keyword.  

[06:00] So, if we have driver golf irons, if we type in  best as a modifier and we click show results,   then we'll have things like best golf balls,  best golf courses near me, you know, queries that  

[06:13] people are actually searching for. So, once we've  filtered this down, we've taken all the different   um queries like from the questions, different  modifiers, manually searching through some,   we're going to end up with a huge list of  keywords. Now, after you've taken these four  

[06:28] million keywords and and narrowed it down to a few  hundred or a few thousand, you have to actually   look at the search results for all of these  queries because you need to figure out if you have   a chance at competing against these competitors  in the search results. And B, you have to assess  

[06:44] search intent. And search intent represents the  reason behind a searchers query. So, can you   actually match the reason for the intent? So, if  we're looking at best golf balls, we're looking   at the search results here. Okay, Reddit's ranking  here. Today's golfer is is pretty authoritative,  

[06:59] too. Uh, Outbound Golf, this is a DR36, so it's  not that authoritative of a website. DR is not   the be all end all of things, but we see Amazon  here. And we can see that the intent overall is  

[07:12] is quite commercial, but we can see that blog  posts are ranking here. So, for a content site,   of course, you can rank with with a blog post. So,  we would decide, yeah, you know what? we will go   for best golf balls and target that. Now imagine  doing that hundreds or thousands of times, right?  

[07:28] Super long process and we're not done here. Then  you have to figure out how these topics all kind   of blend and mesh together like which posts should  be linked to and from and we create this internal   web structure so that search engine crawlers  can come and access all those sites easily  

[07:45] understand which ones should be connected to each  other. So keyword research is just a much bigger   process than just finding keywords. And as I was  building this, I ran into problems pretty much at  

[07:57] every stage. And so I'm going to highlight some  of those for you and how I was able to fix them.   The solutions that that uh AI and I were able to  come together to solve those problems. Okay. So   the first thing that AI did was it had to come  up with seed keywords. And it's super important  

[08:14] that you get your seed keywords right because  everything else in keyword research that comes   after are built around the seed keywords. And so  the first run at this, it came up with seeds like   golf shoe reviews or striicks on versus titleless  golf ball. Like this is super specific. And the  

[08:31] and the problem with this when you use these as  seeds is when you go to generate keyword ideas,   you're not going to come up with any. So in this  case, these two seeds only have 90 keywords.   And if you look at the keyword list, like there's  no volume behind it. There's no demand. So,  

[08:46] it's not about going for like these high ultra  high search volume keywords, but there just isn't   any demand around these. And sure, you can create  content around these, but is it worth the time   and effort? Probably not, right? So, I had to tell  it specifically to go much broader. So instead of  

[09:04] using golf shoe reviews, golf ball reviews, golf  hat reviews, just go with golf and we'll figure   that stuff out later. And then go with instead  of strixon versus titalist golf ball. Like that  

[09:18] is one query that we might actually target,  not a seed. Let's go with titleist different   manufacturers and then go and expand from there.  Now, the problem with this approach though is   that if you're only going for the usual suspects  like golf and putter and putters, everyone else is  

[09:35] doing keyword research that same way. And so, it's  tough to find lowhanging fruit that people aren't   necessarily targeting, but there is still a lot of  demand around those topics. And so, what I had the  

[09:47] AI do is to actually mine competitors keywords  using the HF's top pages report. So it would   find the competitors. It would mine those queries  to see what they're ranking for where there might  

[09:59] be a pattern with certain seed keywords. And so  was able to actually find stuff that you normally   wouldn't be able to find or that wouldn't be  super obvious to find especially in a first   round of keyword research. And so agent A was able  to get the seeds down perfectly and it was able to  

[10:16] go through to the next stage. So to show you an  example, I'm looking at the cluster right now for   golf apparel and shoes. And it was able to group  all of this stuff together. So like golf shoes,   golf shorts. Why is it telling us to go for the  head term? Well, it's because there's a listical  

[10:30] that's ranking there in the top five, right? So  it's actually looking at the search results here   and it's able to determine that. It's able to  look at all these different apparel type stuff,   all these apparel type queries, best  golf sunglasses, and telling us, yeah,  

[10:45] this is probably something that you should go  for as well. And there's so many different things   in here. And obviously, you don't have to create  content around every single topic in the cluster,   but it's meant to be here where you get everything  shortlisted for you. Now the next problem that  

[10:58] actually came is related to this and you can't  see it here but the first round of creating this   keyword research tool. There were some serious  issues with understanding intent. So the way that   AI made this system is based on like a verdict  engine. So it decides whether you should go for  

[11:15] them. So these are queries that fit a golf content  site. There are a group of may where it's like you   could go for it, you don't have to. Um, but yeah,  these are things that you can look up manually  

[11:27] if you want to expand your keyword list further.  There's ones that lack data. So, those ones I just   skip over because we have plenty to go with. These  ones are definitely skips. And so, for example,   Topgolf, you're not going to rank for it. And  even if you do, people probably aren't going to  

[11:42] click it because it's a very much a navigational  search, which it shows right there. And so, in the   first run, this keyword research tool was very,  very bad at identifying intent. It told me that I  

[11:54] should try and rank for Topgolf. Can you rank for  it? Yes, this person does because of an editorial   site. And this is where human overlook is actually  quite important. And so I was able to train the  

[12:06] AI to almost think and act like a human as it  goes through this and make that decision like,   hey, probably not going to make sense to go for  Topgolf or golf or golf courses near me. So yeah,  

[12:18] it was it was waving through branded SERs like  Topgolf or Scotty Cameron and they're saying that   these are winnable content opportunities when  in fact they weren't. The first example that   I actually went with was related to the coffee  niche. So it's like um it was basically waving  

[12:32] through things like espresso machine brands. It  was telling me to go in and rank for like brevel   and I had to correct it and teach it. So look,  it's telling me that the pillar keyword here was   Breville espresso machine. Like with a content  site, you're probably not going to rank for that  

[12:46] one there. it's probably not worth the effort in  that immediate moment. So I was saying, shouldn't   it be something like best espresso machine? So  the thing with AI is that if you ask it to do   something broad, it's going to do something broad  and it's not going to be very good. So that's why  

[13:00] we had to go through this entire process, but  I will tell you the main things that I changed   here. So there were a few major adjustments that I  used to fix this. So first, I added a branded SER   detector. So, if there was a brand that was just  constantly listed in like the top three results,  

[13:16] like if it was telling me to rank for Topgolf, you  know, you would see Topgolf, their homepage, you'd   see their Twitter, their Instagram, and just a  whole bunch of things related to Topgolf. You know   that it's a branded navigational SER, and so it  would know to tell me, let's skip over this one.  

[13:31] Okay. Okay. So, the second thing that I made was  a publisher dominance rule where it would look for   all these mega sites like like Forbes, USA Today,  and if the SERs were completely dominated by them,   it would just kind of skip over it or add it to  the maybe pile. Um, the next thing was able to  

[13:47] identify free tools. So, when it's looking for  keywords, it's actually looking not just for   content keywords like blog posts, but it's also  looking for video topics and it's also looking for   free tool opportunities. And so identifying which  ones are free tools was not as straightforward as  

[14:05] it might seem. And so I had to train the tool to  actually identify which ones were free tools. So   if we look here, I can go to recommended type  and it's has a classification here called free  

[14:18] tool landing page. Let me just go for the goes.  So there are 342 free tool landing pages. Right?   Look at all this. The golf handicap calculator.  We can see that these are all golf handicap  

[14:33] calculators. A golf club length calculator. These  are all tools. Golf swing analyzer. These are all   different tools. Basically, I was going through  this loop of looking through data. So, it would  

[14:46] basically do, you know, the more technical stuff  of coming up with ideas and the logic to make this   work. And I would use kind of my SEO brain to be  like, "This one doesn't seem right because of X,   Y, and Z." And then we were able to come up with  ideas to come to a pretty good solution. So,  

[15:03] and AI will actually go and grade the keywords  one by one. So, yes, there are still some that   we have to filter through, but in my experience,  it's very, very little across three different   niches that I tested. Now, the biggest problem  I ran into was identifying irrelevant keywords,  

[15:20] especially with homonyms. So, this is where AI  is absolute money. So if we go to the overview   report, we look at the executive summary. You can  see here that these are the seed keywords that  

[15:32] were used. So like golf, callaway, tailor made,  these are all brand names, by the way. Puttern,   irons, driver. So you can see like iron can mean  many different things. Driver can also mean many  

[15:46] different things. So, it actually tells us here  every single time it runs through this process,   when it goes through that filtering stage, the  part that I said takes the longest, it's actually  

[15:58] making sure that all the keywords are relevant.  So, it's actually filtering out here 2,456 off-topic keywords. That's crazy. So, for example,  Spark driver, Door Dash driver, Mini driver, the  

[16:13] actress, Baby Driver, the movie. It's constantly  looking at these these keywords and making sure   that you don't have to actually filter them out.  And I think that was the best part is that AI has   all this knowledge already. So as it goes through  it, it's able to say mini driver, nah, not nothing  

[16:30] to do with that. And it'll just take out anything  that's related to that. So every single keyword   runs through AI and it removes the keywords  that are irrelevant to the niche of golf. Now,   not everything was bad. There were some things  that were actually great uh out of the box. And so  

[16:46] here, one of those things was content briefs. So  if we look at like I don't know, best golf pants,   this is one that I generated. Um when I click  those buttons to to generate content briefs,   it'll do things like this. It'll show you the  target keyword. It'll tell you about the SER.  

[17:01] So you know, your content writer can go in there  with, you know, some basic knowledge and know who   they're competing against, the types of pages.  It actually goes and it reads the different pages   that are ranking in the top 10 and it's able to  identify weaknesses, areas that you can capitalize  

[17:15] on to create better content. So, it says here that  for best golf pants there's a weak editorial that   there's a weak editorial landscape and only one  true listical which is kind of interesting to me.  

[17:28] And we can see here that the KD is one the keyword  difficulty that there's weak top results that the   single editorial content is a DR51. It's doing the  analysis for you kind of of what you would do when   you look through everything, but you don't have to  look through all these SERs at once. You just go  

[17:43] through them one by one and it's all listed here.  It'll give you suggestions for titles. So, this is   telling about intent. It's telling you to create  a listical here clearly. Uh the metad description,   metad description is not so important anymore.  It'll tell you the URL slug to use a word count  

[17:58] an exact word count here. Um, but it's, you know,  giving you the median for the people who do care   about word counts or people who are paying other  writers to hit certain word counts. And then it'll  

[18:11] do the entire outline for you. And the outline  is actually decent. So, it's already looked at   the different listicles that are ranking and it's  providing suggestions to you based on that. So,   it'll tell you, you know, we've seen affiliate  sites do this before is they always have this  

[18:26] thing of like how I tested these golf pins or  whatever it is. I think that's actually quite   good for the reader as well to know that you have  actually reviewed these, you've touched them,   you've worn them and tested them. So, we want to  give people things at a glance because especially  

[18:41] with blog readers today, everyone is skimming.  So, make sure that it's skimable. We provide   all that info for them. Do you have to label  them like this? No. You can give your honest  

[18:53] reviews like that, but if you're truly going  for like content site type thing, you know,   you at least have seeds and suggestions of what  you might want to do. This part is pretty cool.   It tells you key entities and topics to cover  basically to have, you know, full relevance  

[19:08] and topical coverage here. So, you know, Lulle  Lemon, um, Travis Matthew, Bonobos, Nike, Adidas,   etc. gives you an FAQ and says that it's mine from  Reddit threads in the SER. I did not know this,  

[19:20] but I think that is pretty cool is that it's  picking up things like people also ask questions   and it's bringing all of this together. And  this is where the money is, right? The in the   internal linking part is it's telling you that  from the target page, so the golf apparel shoes,  

[19:35] which the bigger part of the hub, it's telling  you to link from that page, it says this is a   spoke page. So this is one of those branches,  right? It links up to the pillar and across to   relevant siblings. So, it's telling you to link  from this page. Uh, tells you the anchor text. It  

[19:51] tells you why you should do this. And so, all this  stuff is super super important. And I think that,   uh, you know, most content tools are are  missing these kinds of features. Again,   you need to have some kind of SEO knowledge to be  able to make good decisions in this. Take things  

[20:10] with a grain of salt. Now, another thing that it  did well out of the box is the hub and spoke map.   So right now I'm just looking at, you know, the  go keywords here and there are there is a lot of   stuff going on here. So let's say that I wanted to  just look at the quick ones. Then we can actually  

[20:24] see the clusters. So we can see the different  pages here. Um like what makes a golf cart street   legal, how to test golf cart batteries. You can  tell that these are going to be low competition,  

[20:37] but these are meant to be quick wins. You  basically have options is my point here. So   the final result, you know, we went from entering  a niche. I entered just golf. It came up with a   seeds and went through over 10,000 keywords,  probably closer to like 400,000, and it would  

[20:53] bring it down to like 10K. It would go through a  relevance filter. It would cluster stuff. It would   pull all the live SERs. It would provide all  the classifications and do the competitor gap,  

[21:05] create the hub and spoke map. It would do per  keyword briefs which is on demand because we're   not going to generate all of them obviously  because that's a waste of API tokens. And you   have the option for CSV PDF exports. Now let's go  back and see if our house plans thing is done. Oh,  

[21:22] it's done. Literally done right then and  there. So gez, it went through. In fact,   11,800 keywords. And this is on house plants. And  for the record, I know nothing about house plants.   So leave in the comments if this did a terrible  job. We'll go look through some of the keywords  

[21:37] and see how well it did. So, the overview here, it  found over 6,000 queries related to house plants   that we could potentially go after. And this is  going to be a real test because with house plants,   I would assume that the majority of queries are  going to be commercial where, you know, you have  

[21:52] to go and there's going to be e-commerce product  category pages, service pages. Um, but yeah, so   we can see here these are the seeds that it came  up with like a monstera paos. I don't know how  

[22:04] to say that. orchid, snake plant, peace lily. So,  it went out and it found the seeds that are like   popular house plants. I don't know about that.  I've heard of I've heard of these some of these,  

[22:16] but that's it. Top content opportunities, Mona,  varieties, and care. Okay, that's cool and helpful   to know. Share a voice. This is interesting.  the share of voice in this niche. The majority   for house plant care advice and it's giving us a  recommended action plan. So ship the Mona pillar  

[22:37] first. It has the highest opportunity score and  supports a deep cluster of variety of listicles,   care how-tos, and problems solving guides. Cool.  Let's take a look. So we're just looking at go   keywords here and let's look at the clusters. So,  it's said to look at the Mona varieties in care  

[22:54] uh or Monstera toxic to cats. This is considered  a quick win and a go. So, let's let's generate a   brief on that for now. So, basically, we have all  the clusters here. We have the competitor gap and   we can see the the websites that are getting the  most traffic for these queries. So, the spruce  

[23:12] that makes sense. Lively root sounds like it makes  sense. Uh and the hub and spoke map. This one I'm   a little nervous to see. I'm not looking yet. But  yeah, that's crazy. Oh man, like these these hubs   and spokes. You know what? Because it's limited to  I limited to I think 15 to 20 for medium searches,  

[23:29] it's actually quite reasonable to look at. And  being someone who knows almost nothing about house   plants. I can come here and I can see how some  of these should be connected. Saying that there   should be reciprocal links here because this one  links to here. So this one on alocasia varieties  

[23:46] and care should link to airplane care and vice  versa. We can see the different ones that content   that we should create. We can see the supporting  content here. This is all go content. But this is   all done for me and I think that it's done a very  very good job. Uh the content briefs they're all  

[24:03] done uh generating now too. So let's look at the  one for cats. We have the title. So are monstera   plants toxic to cats? What every owner needs to  know. Vet reviewed. Okay. Well, that's not true.   But, you know, we could get a vet to do that. I'm  sure it's done the competitor analysis for you.  

[24:19] Like, I like I'm actually genuinely impressed  with this as I'm as I'm going through this. I   don't know how accurate this stuff is and how how  factual it is. I'm not a vet either, but you can   see here what to do if your cat shoot a monster  leaf step by step. It's telling you everything to  

[24:36] do. Call your vet or ASPCA poison control. Will my  cat is just answering questions really of someone   who's searching for this. But from what I'm seeing  here, I can see that it does a lot of really  

[24:52] really good work for us. And all I did was type  in one word, house plants, and it came up with all   of this, which is super cool. Now, the the crazy  thing about this all is in agent A, if I wanted  

[25:06] to make a different change to this, let's like I  don't like the way that let's go back to our house   plans. Let's say that I wanted to have I didn't  want to have that limit of like 15 to 20 clusters,   but I wanted to have as many clusters as as  needed. It would go and do potentially an infinite  

[25:22] number of those, right? Um the hub and spoke map.  Let's say I don't like the colors or when I click   on one of these, let's say I want there to be  like a little box that pops up and I can generate   the content brief directly from the hub and spoke  map. All I have to do is go in here, type that up,  

[25:37] enter it, and it'll make those changes. Now,  if you want this exact keyword research tool,   you can build it yourself using the steps that I  walk through. You know how to solve the problems,   too. Or if you've got an agent a workspace,  you can oneclick install it from the app  

[25:51] store. Technically two clicks uh and have  it installed directly into your workspace.   And if there's anything that you want me to  build for you, tell me in the comments below.