Research
Last updated
Last updated
Research is an AI-driven feature
While we strive to ensure that citations and links are as accurate as possible, this is still an AI-driven feature and can still experience some hallucinations and inaccuracies.
It is your responsibility to ensure that the information is factual and that the external links are correct - you should always be checking and vetting every article you generate.
Automatic mode allows you to scrape the SERPs automatically based on the topics and keywords in your articles. This will retrieve the latest and highest ranking sources of information on a topic, providing valuable context for the writer and extract relevant facts and figures to use in your article.
Additionally, you can add specific domains to exclude from research, so you can avoid things like competitors, reddit and so on.
Machined doesn't simply check the SERPs for the keyword of the article - it uses AI to analyse the main search results and then performs a further set of searches that are more targeted to the topic at hand. At minimum, 5 separate searches are performed for each article (much deeper than most AI writers that provide "real-time" data).
Semi-automatic mode lets you provide up to 5 urls to scrape for research, giving you much greater control of the context and data that shows up in your article. This could be a list of specific high-quality articles, including your own pages.
Our AI is smart, so it will only include research from pages that are truly relevant to the topic and keyword. With this in mind please only supply relevant webpages to scrape, or you may waste time and money scraping pages that are never used.
Both automatic and semi-automatic modes can be used together 🚀
For all researched articles, you have the option to include references and citations to the content that the research originates from.
References are explicit callouts of information from the research, they normally take the form:
"A study revealed in American Scientist ..."
"According to HealthLine ..."
"Data from CNN"
At the moment references are always styled as inline links and there is no density control.
Citations are attributions of knowledge to a particular source, usually in the form:
"... dark chocolate is the healthiest (Healthline)."
"... that 78% of dark chocolates contain at least some milk (Lactose Journal)."
There are currently two styles to choose from, Inline and Footnote.
The Inline style is the standard style and is shown above.
The Footnote style introduces a new "References" section at the end of the article, moves all physical links there, then replaces the inline links with local footnote references.
Effectively, footnotes allow you to have many more citations without being penalised for using the same external link multiple times, since the link only appears at the bottom of the article.
We recommend you use either Inline style with Standard density, or Footnote style with High density
Research is still experimental and we are yet to optimize the process and the prompts - however the current average costs are below and will improve over time...
| Model | Research cost (avg) | Article cost (avg) | Time to write (avg) |
|---|---|---|---|
GPT 3.5 | $0 | $0.10 | 20 seconds |
GPT 3.5 + research | $0.08 | $0.25 | 30 seconds |
GPT 4 | $0 | $0.85 | 50 seconds |
GPT-4 + research | $0.08 | $2 | 60 seconds |
GPT-4 Turbo | $0 | $0.21 | 40 seconds |
GPT-4 Turbo + research | $0.08 | $0.48 | 50 seconds |
Research results in three types of additional costs
Cost of search
Cost of research analysis
Cost of writing articles with research
The costs of using the underlying Search APIs to search for research material are covered under your Machined subscription, although this may change in the future.
There will be costs associated with analysing research content; to summarising and catalogue the content effectively. The exact costs for this can vary based on the amount of text that is pulled in for each search result but it ranges from a few thousand to tens of thousands of AI tokens.
We always use gpt-3.5-turbo and gpt-3.5-turbo-16k to do this work.
The average cost per research run is $0.08
There will also be extra costs incurred when writing the article itself because we need to feed snippets and summarised versions of the research to the AI writer. These costs may also increase based on whether you decide to include citations or not.
The impact on costs greatly depend on the model used to write the article, and you will especially see a large cost increase if you are using GPT-4.
Refer to table above to understand better how research impacts cost.
Writing time will increase by roughly 10-30 seconds, depending on factors like how quickly APIs respond.
