What does AI say about you? Is asking AI to research you the new Googling yourself?
I asked all of the major AI large language models the same question, and these are the results.
For all of them, I chose the in-depth research option where available. I left them all at the default choice of model, as for most users, choosing between models is far too confusing.
The question was:
“Please provide a detailed report on Stuart Bruce who works in PR and isn’t the music producer.”
Some models ask questions to clarify what you need. My answers were always everything, everywhere. The “isn’t the music producer” is to ensure they don’t include references to Stuart Bruce, the music recording engineer who engineered and mixed Band Aid’s ‘Do They Know it’s Chrismas’.
All of the answers were positive and most were highly complementary. In fact, some of the language about achievements and accomplishments phrased them far ‘better’ than some of my own descriptions of them. They don’t appear to suffer from typical British understatement, so can be used to give inspiration if you’re trying to update your biography or LinkedIn profile.
The exception to the glowing descriptions was Perplexity, which provided far more neutral descriptions.
What does AI say about you? – Stuart Bruce
These are the individual results from Perplexity, Copilot, ChatGPT, Grok, Claude and ChatGPT.
Perplexity
Paid pro account
Best for quick, accurate research.
- 1,314 words
- 2 minor mistakes
- 19 sources
Review
Perplexity was the most current, as it clearly recognised my primary focus is now on AI, saying:
“Bruce is recognised as a leading expert in AI and technology for public relations.”
It was also the only one to correctly identify that I now sit on the PRCA PR and Communications Board; all the others provided historic information about me sitting on the PRCA PR Council.
Overall, Perplexity succinctly covered most of the relevant information in a reasonably neutral tone of voice. It was also quick, making it suitable for quick research overviews with a reasonable confidence of it being both accurate and current.
Accuracy
Perplexity’s two minor mistakes were referring to the 10-year-old Barcelona Principles 2.0, rather than the two more recent versions, and having the old ‘A PR Guy’s Musings’ name for my blog instead of PR Futurist.
Sources
Perplexity relied heavily on LinkedIn and owned media as sources. It also cited podcasts and YouTube.
Perplexity had 1-18 source links throughout its answer, but then bizarrely also had the number 28 as a source link. Its list of sources was 87, but this included several that were the wrong Stuart Bruce, but it didn’t appear to use these in its answers.
Link to Perplexity report
Copilot
Paid Microsoft 365 Copilot business account
Best for comprehensive, detailed report.
- 4,358 words
- 2 minor mistakes
- 7 sources
Review
Copilot was the most comprehensive report. It had the most logical structure and was the most attractively laid out with information panels and a timeline. The Copilot report starts with a useful introduction summarising each section.
The ‘Current Projects’ section was the most useful as it correctly summarised what my current focus is. It included an entire section on Purposeful Relations and what we do.
Its groupings for ‘Industry Contributions’ were the most logical with sections on:
- Advancing Professional Standards
- Education and Mentorship
- Thought Leadership and Knowledge Sharing
- Promoting Innovation in PR Practice
- Global Dialogue and Networking
- Advocacy for Corporate Purpose and Ethics
Accuracy
Copilot made two minor errors: stating there are 500 chartered practitioners instead of 861, and incorrectly referring to the PRCA council instead of the current PRCA PR and Communication Board.
Sources
Copilot cited seven sources, which were all owned except for one earned media source – PRWeek.
Copilot report
I can’t provide a direct link to Copilot’s report as the business version doesn’t yet allow you to share pages externally so I’ve had to embed it as PDF.
Gemini
Free account
High risk as made serious mistakes or hallucinations
- 5,015 words
- 3 serious mistakes and 4 minor mistakes
- 7 sources
Gemini provided the longest research report. This is probably because it is very verbose using entire sentences that don’t add anything substantive. It also had sections repeating itself.
Gemini’s writing style is also the most ‘clunky’, using formal language such as “Mr Bruce”. It would need the most editing to make it read more naturally.
Gemini was also the most opinionated. For example, when it talked about my experience as director of comms for “two senior cabinet-rank UK politicians” its opinion was:
“These roles underscore his versatility and ability to operate effectively in high-stakes, politically sensitive environments, adding a unique dimension to his overall expertise.”
It then had an entire paragraph where it expressed opinions as to what I had learned from my “diverse professional background, encompassing agency leadership, in-house corporate roles, and high-level political communications.” Some of its speculation was true, other parts not.
The entire report is riddled with Gemini expressing opinions, rather than just describing what it knows.
Accuracy
A big concern is that Gemini was the only one to make serious mistakes or hallucinations. Gemini’s first serious mistake was in a table of my career milestones where it claimed I’d led an agency in the early 90s (when I was a mere account executive!) rather than the late 00s.
An even more serious error is that it created a table of my client experience where it had a column for “Nature of engagement (if specified).” Of the 19 clients it listed, it got 11 wrong. It simply made up what I’d done for them. With some, it wouldn’t have been able to find the answer as while I list them as clients I’ve never said what I did. With others, the correct information is available on owned media.
Sources
Gemini cited six sources – five owned and one earned.
Link to Gemini report
Grok
Free account
Doesn’t cite sources, so use with caution
- 742 words
- 2 minor mistakes
- 0 sources
Review
Grok was the most surprising, as I had low expectations. It actually rivaled Perplexity in creating a quick, accurate report. It might have been short, but it included all of the relevant information. It was the only one to include fun or personal information, such as:
“He once competed in the Junior Working Hunter Pony Championships, a light-hearted anecdote he occasionally shares.”
This information is easily available online, so it’s disappointing other reports didn’t include it.
Accuracy
Just two very minor mistakes. One year out for when I started my career and a reference to the PRCA council (a mistake most made).
Sources
No sources cited, which means it can’t be trusted to research something you don’t know anything about.
Link to Grok report
Claude
Free account
Better at writing than it is at research?
- 423 words
- No mistakes
- 3 sources
Review
Not much to say about Claude as its report was so insubstantial. On a positive note, it was probably the ‘best’ written. It was simply short and factual.
Accuracy
Claude was the only one not to make any errors. But that’s probably because it contained the least detail.
Sources
Three sources, all owned.
Link to Claude report
ChatGPT
Paid pro account
High risk as made serious mistakes or hallucinations
- 3,255 words
- 2 serious mistakes, 8 minor mistakes
- 31 sources
Review
When I started this exercise, my hunch was that ChatGPT would be the winner. That’s not the case. It used the most diverse range of sources, which meant it included some interesting additional information, although none of it made a substantial difference. But this was balanced by the fact that it included serious mistakes. One of these was with a citation, so could be checked, but you would have to know it needed checking. The other didn’t include a citation, so would be hard to check.
This makes it hard to recommend ChatGPT as, while it was the best in some respects, it was one of the only two that made serious hallucinations.
Overall, I don’t feel ChatGPT’s serious mistakes were as bad as Gemini’s serious mistakes.
Accuracy
Both of ChatGPT’s serious mistakes are probably because it was trying to be over-complimentary and made stuff up about my role in things I’m interested in and comment on, but actually had no role in. It claims I “contributed to industry research initiatives like the Government Communication Service’s “Performance with Purpose” strategy review.” I didn’t, I simply shared it and use it. It also said “he was part of discussions on initiatives like the Barcelona Principles for PR measurement.” I’ve never been part of the working group discussing and writing the proposals. What I have done is provide private feedback to people on the group and use the Barcelona Principles extensively in my work.
Some of its minor mistakes were the same as others (PRCA council), but others were just odd. Such as saying the PR Futurist newsletter is “hosted on Substack/beehiiv” when it is on Curated.
Sources
ChatGPT used the broadest range of sources, including earned media like PRovoke and Communicate. It was also the only one to cite our PR Futurist newsletter as a source, rather than just saying I publish it. I’d have hoped this broader range of sources would make it more accurate, but sadly that wasn’t the case.
Link to ChatGPT report
Recommendation
While this was mainly a fun exercise, it does highlight the need to review all of my owned assets to ensure they are consistent and optimised for GEO – or generative AI engine optimisation.
Methodology
This exercise wasn’t intended to be a robust comparison the research capabilities of different large language models. Therefore, the methodology wasn’t robust enough to draw definite conclusions.
One issue is I didn’t compare like for like, as some models were paid versions and others free versions.
One area where the data might be particularly inaccurate is the number of citations. Different AIs provide the information in different ways. For example, Perplexity had 1-18 links throughout its answer, but then bizarrely also had 28 as one of the links. Its list of sources was 87, but this included several that were the wrong Stuart Bruce, but it didn’t appear to use these in its answers.
