This document provides an overview of AI-generated content and tips for using it effectively. It discusses how large language models (LLMs) work by predicting the next word based on statistical patterns in training data, but don't truly understand language. While AI content requires human oversight for accuracy, many businesses are using it to generate drafts or ideas. The document recommends prompt engineering to guide AI responses and tips like having AI suggest content topics or parts of drafts while leaving final revisions to human editors.
3. ● The Basics of LLMs and Generative AI
● What This Means for You
● Tips for Your Content
Agenda
4. Why should we care about AI-generated content?
42%
Percentage of businesses
who are already using AI
for long-form written
content
Source: Forbes Advisor
70%
Percentage of
businesses who think
that GPT will help
generate content quickly
Source: Forbes Advisor
6. Generative AI has now reached the point where it
sounds human-like…but is it?
7. LLMs sound sentient because they have been trained on
a massive volume of internet text data
Internet text Statistical model LLM output
8. OpenAI further refines the model using reinforcement
learning with human feedback
Source: InstructGPT paper
9. Essentially, LLMs guess the best next word in every
sentence, based on their training data
Input
The first president of the
United States was George…
10. Essentially, LLMs guess the best next word in every
sentence, based on their training data
Input
The first president of the
United States was George…
11. Essentially, LLMs guess the best next word in every
sentence, based on their training data
Input
The first president of the
United States was George…
Output
…Washington
12. Essentially, LLMs guess the best next word in every
sentence, based on their training data
Input
The first president of the
United States was George…
Output
…Washington
And they do this really, really fast
13.
14. While LLMs know the statistical structure of language,
they don’t understand meaning
15. While LLMs know the statistical structure of language,
they don’t understand meaning
16. “Lies, damn lies, and statistics”
Source: OpenAI GPT-4 Research
Even the best
generative AI models
are only about 80%
factually accurate
18. Good News: Google doesn’t oppose AI-generated
content
● As long as your content demonstrates E-E-A-T and meets
the needs of the searcher, Google will not penalize you for
using AI to assist with content generation.
19. “
Google Search’s Guidance about AI Generated Content
Source
AI has the ability to power new levels of
expression and creativity, and to serve as a
critical tool to help people create great
content for the web.
20. Be wary of accuracy and originality when using AI
22. Your Starting Point: Prompt Engineering
The practice of designing and refining the
inputs provided to an LLM to produce the
optimal result.
23. The best prompts take advantage of your knowledge of
how a specific LLM has been trained
Source: InstructGPT paper
Tips from OpenAI:
● Be specific, descriptive, and
detailed
● Provide examples
● Use leading words to nudge
the model towards a desired
pattern
● Give clear instructions
Source: OpenAI knowledge base
24. Spend time experimenting with prompt engineering to
generate the most relevant content
● If at first you don’t succeed, try, try again. Be precise about your intent.
○ “Is ‘make sure’ a keyword that a user is likely to type into a search engine?” vs “How
likely is a user to type ‘make sure’ into a search engine as their entire query when
looking for information about "How to choose the right VPN for you - 9 key things to
look out for"?”
● Specify the answer format. These are some examples for creating reusable prompts that
can consistently process data.
○ “Answer in 1 word.”
○ “Create a table where the first column is a keyword and the second column indicates
whether a user is likely to type the keyword into a search engine as their entire query
when looking for information about "How to choose the right VPN for you - 9 key things
to look out for" in a score from 1 to 10. The keywords in the first column should be
"new tab", "make sure", "view deal", "image credit", "tab view", "vpn buying guide" and
"how to set up a vpn"”
25. Tip 1: Brainstorm new content ideas
Benefits:
● Provides insight into common
themes
● Leaves room for your own
voice, tone, style, and
expertise
● Avoids factual errors or AI
hallucinations
26. Tip 2: Write parts of first drafts
Benefits:
● Requesting individual parts
of a bigger first draft can give
you strong, highly targeted
content to incorporate into a
larger flow
● AI content generators often
follow a formulaic content
structure that can be time-
consuming to revise
27. Tip 3: Revise existing content
Benefits:
● Can allow you to easily
target specific audiences or
SERP elements
● Retains key elements of your
original content
28. Tip 4: Review content and provide feedback
Benefits:
● Given clear constraints, can
provide direction for the most
impactful revisions
● Especially useful when you
are unsure about how to
improve your content for
specific keywords or PAA
questions