Insights

AI Content & SEO Workflow: Mitigate the risk of AI Inaccuracies with P-A-R-S for E-E-A-T

The Great SEO Reset is here, and many will be left behind due to fear of the unknown

Dramatic? Yes. Accurate? Also yes.

Wil's 2023 MozCon presentation highlighted the idea of The Great Reset: Yet another memorial of SEO brought on by the disruption of AI tools.

As many have pointed out, SEO is alive and well. But that doesn't mean your current search market share is safe.

As AI continues to disrupt all walks of life, we're confident a reset will take place that shakes up the current field of winners and losers in search. Those who come out on top will be the ones leveraging and embracing new ways to work and new discovery paths, all brought on by AI.

This post will highlight how we can embrace new ways to work.

Note: Throughout this post we'll be exclusively referencing ChatGPT-4. For simplicity and narrative flow, we'll consistently use the term ChatGPT.

The race has begun and you may already be behind

Several clients have told us their organizations have chosen to block certain AI tools from use completely - the AI gate keepers have arrived.

This aligns with BlackBerry's recent survey of 2,000 IT decision-makers which found 75% of organizations globally have banned or are considering banning ChatGPT and other generative AI apps.

Of those clients surveyed by Seer who said AI accuracy is a concern, 62% are in Senior to Executive marketing roles. This quote sums up their concerns well:

"How can we maintain accuracy and integrity in our messaging [leveraging AI Tools]"

As marketers, we rely on accuracy and brand consistency to help stitch together great experiences for our users. This is of critical importance, and no efficiency gains are worth the cost of risking the brand's integrity.

That said, if you aren't actively identifying how you can do both (gain efficiency and protect the brand's integrity) with AI, you're behind. There are competitors moving full speed ahead with AI generated content, AI driven analysis, and even direct AI integration into their products. 

They're moving faster and working smarter while you ping your legal team for the 8th time on their input. Let us help by demystifying Generative AI with a concept everyone can understand: a human powered workflow.

Mitigate the risk of AI inaccuracies with P-A-R-S for E-E-A-T

Just what you've been waiting for - more search marketing acronyms!

Brief History of E-E-A-T

Previously, the north star for quality could be summed up by E-A-T:

Expertise

Authoritativeness

Trustworthiness

In December 2022, Google released an update to their quality rater guidelines to add Experience as a new core tenant.

In my opinion, Generative AI represents the most exciting opportunity in SEO since SEO friendly site crawlers like Screaming Frog were created. 

But, with great power comes great responsibility. We need to ensure our approach stays true to our north star for all optimization efforts by demonstrating Experience, Expertise, Authoritativeness, and Trustworthiness.

Simple prompts that rely on the existing training set data are not going to cut it.

To truly incorporate E-E-A-T into your ChatGPT process, you'll need one more acronym: P-A-R-S.

What is P-A-R-S?

P-A-R-S is a AI content and SEO workflow that ensures we can meet the standards of E-E-A-T while gaining AI efficiencies. The acronym stands for:

Populate

Activate

Retrieve

Scrutinize

This approach can be leveraged for a wide variety of SEO activities. It can also serve as part of the risk mitigation plan to help you get your stakeholders on board to use Generative AI.

Use P-A-R-S as the base of all of your ChatGPT workflows to ensure:

  1. Your prompts will create high quality outputs
  2. Your review process will mitigate any marginal risk

Here is how we’ll be approaching this at Seer:

Breakdown of P-A-R-S to maximize E-E-A-T

Step

Action & Description

Populate

Populate unique inputs into ChatGPT. This ensures you're leveraging more than the data the model was trained on. 

What unique inputs, you ask? This is the core way you'll integrate E-E-A-T into your approach. Some examples:

  • Transcripts from conversations with SMEs can infuse experience into your dataset.
  • Exports of People Also Ask data can help you ensure the expertise of your content is represented by speaking to the questions your audience has.
  • Links to trusted sources of deep technical information, like our Guide to Screaming Frog, can infuse trustworthiness

Import data directly via Code Interpreter, or instruct ChatGPT to visit a public webpage to learn more about a topic via a plugin like WebPilot.

Activate

Activate your data + the power of ChatGPT with the right sequence of prompts and feedback.

Learn more than the right prompts to use, and instead focus on the anatomy of successful prompting and feedback loops.

Retrieve

Retrieve the output from ChatGPT by exporting into your desired file type using Code Interpreter.

In order to avoid too much manual intervention between retrieval and scrutiny, your first two steps need to produce high quality outputs.

Scrutinize

Review your output with a variety of pre-set acceptance criteria via a variety of individuals.

If you're agency-side you likely already operate with Acceptance Criteria or SOPs. If you're with a smaller team, you may need to create these.

This workflow can be leveraged for all tactical execution, from metadata optimizations to content outlines. While the overall process isn't tremendously different than how you've approached tactical execution in the past, you're introducing many new steps and tools for your team to use.

Change management can be hard, especially for larger organizations. But, for those willing to do the hard work and have the hard conversations, the gains will be tremendous.

Here's more detail on each step.

Example of P-A-R-S

Populate

Let's start by populating the dataset with additional details.

ChatGPT-4 is an incredibly robust LLM and we're optimistic for how much better ChatGPT-5 will be.

But at the end of the day, it's a generalist. We're specialists. Our experience is robust and unique. Feeding that experience into ChatGPT is the first step towards leveraging P-A-R-S for E-E-A-T.

Work with ChatGPT the same way you'd work with a new hire.

Share training documentation or helpful links you have bookmarked on the web.

We'll illustrate the process with a very simple example. Let's say you want to add schema markup to a page. You can approach one of two ways:

1. You can assume ChatGPT knows everything it needs to know about schema markup, and ask:

2. Or, you can populate ChatGPT's dataset with additional information:

Activate

Now, let's activate on that expanded data set with the right prompt.

Once you’ve uploaded the relevant data in a way that ChatGPT understands, you can activate by prompting ChatGPT with your desired output.

For more information on how to write great prompts:

Here are some basic considerations from the Harvard Business Review post, plus my own addition:

      • Diagnose the problem. Make sure ChatGPT understands your main objective
      • Break down the problem functionally: If your prompt requires multiple steps, consider the logical breakdown for how ChatGPT can approach it. This includes giving feedback on what they did right and what you want to see improved.
      • Reframe the problem: Ensuring ChatGPT has the context needed to fully understand the task is critical, so it may help to provide different angles of the same view. For more simple tasks, this may be overkill.
      • Outline the constraints: ChatGPT has to understand what success looks like, so feeding it acceptance criteria is critical to the final output. For example, if there are “must include” vs “nice to have” considerations, make sure that information is clearly organized in the Populate stage.
      • Pro Tip: Ask for more. In my experience, you can engineer the perfect prompt, but ChatGPT won't ever leave it all out on the field. Follow up prompts that essentially ask "what else?" can help to round out your output.

Let's return to our example to share how this manifests into the exact prompt.

The addition of populated data and more thought around how we activated on the prompt is the difference between an output that's OK and useable:

1

To an output that's robust:

2

Retrieve

Next we'll retrieve the output in your desired format

If you were successful in the first two steps, all you’ll need to do here is copy and paste your output.

Keep in mind there are a variety of ways ChatGPT can export it's findings for you.

Using the example of creating schema markup for Wil's bio page, we can ask ChatGPT to render it's recommendations in JSON-LD format.

Code Interpreter supports a variety of file types include text, document, image, audio, video, data, code, markup, pdf, and database.

Scrutinize

Scrutinizing the out needs to be done using a consistent and rigorous approach.

Our goal is to end up with the highest possible quality work product, and review is a critical step in that process.

As such, we’ll want to scrutinize everything:

      • Ensure the output is grammatically correct
      • Ensure the output is informationally accurate
      • Ensure the output aligns with the brand voice and tone, if working with content
      • Ensure the output represents the SME accurately and conveys their experience and expertise
      • Ensure the output is unique

In the example we used to illustrate the P-A-R-S process, we would use the following review process:

  • Wil would review for accuracy
  • A member of our Technical SEO team would review the schema markup to ensure it validates

Remember, this process can be leveraged for all types of SEO activities.

Time consuming tasks like content briefs and content audits could become more efficient using the same approach. That said, the scrutinize step for that work would need to be more robust.  For example, the output is reviewed:

      • By an SEO practitioner against formal acceptance criteria
      • By a copy editor or run through a copy editing tool
      • By an SME to confirm alignment with the experience and expertise they supplied
      • By an SME or Account Person to confirm it’s aligned with brand guidelines
      • By a plagiarism checker tool

The scrutiny may feel arduous, but this is also a critical part of optimizing your inputs in the P-A-R stages to improve further iterations.

You may now be wondering if all of these steps are really going to save you time. First, I’d venture that at the very least they won’t cost you additional time, unless the result is also a higher quality work product. Most of these checks and balances should already be a part of your process, so the additional time will be minimal. Additionally, it’s not hard to imagine a world where ChatGPT or other LLMs advance to a point in which they can be trusted to take on some of these steps.

Where do we go from here?

We previously shared guidance for how to make a case for how we can safely leverage Generative AI within our Content and SEO process. Leverage these workflows to help further that case to those concerned about the risks of leveraging Generative AI. 

Stay tuned as we leverage the P-A-R-S process for more SEO and Content activities!

 

We love helping marketers like you.

Sign up for our newsletter to receive updates and more:

Alisa Scharf
Alisa Scharf
VP, SEO + Generative AI