Choosing Inputs

Choosing what to write on, in an AI-first world
This page sums up a few bits of advice that I like to give to SEOs and site owners when it comes to choosing inputs (keywords, titles, et cetera) on Byword.

Volume & difficulty

If you read the last page, you'll understand that tools like Byword open up possibilities to rank on search terms that are too low-volume to make sense when you have to pay a human writer.
With that in mind, when doing keyword research for AI content you should be far less sensitive to volume estimates than you would normally be. It's completely fine for a search term to have one or two hundred searches a month. Even if that only translates to a couple of clicks a day, that quickly adds up when multiplied by hundreds or thousands of articles.

Formats & coding

When analysing search volume reports, it can often be good to practice something called inductive coding. You can read more about this here, but this basics of it is as follows:
  • Scan a search report (from a tool like ahrefs or Semrush) for keywords that differ from one another in one constituent part. Examples of such keywords could be: can cats eat carrots can cats eat chicken
  • Build a coded keyword structure from these keywords. In the case of the example above, this would be can cats eat {food}.
  • Populate a list of items that can be fed into your coding. In the example above, this would be a list of foods which would make sense in the context of an article about cats eating that food.
  • Build out all the keywords/titles in Excel or Google Sheets, and pass that into Byword.
This makes particular sense in the context of AI content production because tools like Byword can easily power through hundreds of articles in this sort of format. This allows you to build topical authority, and produce the sort of content that it'd take weeks (if not months) for a human writer to produce.
The example above used a 1-dimensional keyword format (meaning it had one variable in its coding). 2- (and even 3-) dimensional formats also work well in the context of AI content production; you can read more about them in the page on N-Dimensional SEO.

Content type

If you've played around with Byword already, you'll notice that it performs differently on different types of content.
Byword is at its best when it's writing on factual, informative, educational content. This means it tends to work well on question-based keywords, or really any keyword where the searcher is clearly trying to find something out
Byword tends to struggle with non-factual content. This includes opinion pieces, reviews, or anything requiring a large degree of subjectivity.

Training data

Byword's algorithms are built upon a series of language model prompts. This means that Byword's knowledge is limited to the knowledge of those language models, which is a reflection of their training data.
This means that Byword has a couple of limitations:
  1. 1.
    It can struggle when writing about recent events or phenomena. This is particularly apparent when writing about events that have taken place within the last year.
  2. 2.
    The more that has been written historically about a topic, the better Byword will be able to write about it. Byword is unlikely to be able to write anything high quality about your local neighborhood sports team, but it can write much better content about well-known global sports teams.