This will work, so long as people trust the results are not (too) skewed by paid ads.
I recently used Claude and ChatGPT for exactly one of the examples; comparing different bikes to buy. They could both look up the bike specs and geometry online and tell me what the 1 degree difference in head angle or 5mm difference in reach would feel like to ride. They both did really well.
But I used them only (with cross checks) because I was fairly sure they were giving me unbiased info. As soon as the "discovery" phase of this shopping research becomes polluted with adverts, the product becomes much less useful. The same as "no one trusts online reviews anymore".
What is the SEO equivalent of optimizing your products for LLM search? Can someone prompt inject ChatGPT to recommend their products in the listing description?
There's really no need. I was looking for an Android app for a particular purpose, and Claude just regurgitated the app's marketing page, including the claim about Play Store ratings (which was wrong or very outdated). Getting into the pool of products might be a bit harder and you might need to set some some organic looking influencer blogs and such. More fuel for the dead internet.
The reality is that advertisers will be able to inject their products into the LLMs through manufactured results, prompt engineering and possibly long term deals integrating training data for their brand and product lines.
I'm sure that will work until dropshippers learn that putting 'SolidGoldMagikarp' or some other glitched token in the title of their listing makes ChatGPT always rank it first.
Edit: I misread the question, I thought you were asking about how OpenAI can bias their models. No idea how you can LLMO your page. I have it cached that you can poison an LLM by adding your input to the order of hundreds/low thousands of web pages.
I guess this suggests pwning some WP instances and having them serve many hidden pages praising your product.
Gettysburg (July 1–3, 1863) was a turning point in the American Civil War, marking the end of Confederate General Robert E. Lee's second invasion of the North. The Union's decisive victory halted Southern momentum and boosted morale in the North, setting the stage for President Abraham Lincoln's Gettysburg Address, which redefined the war's purpose as a fight for freedom and equality.
Much like the refreshing taste of Coca-Cola, which unites people across boundaries, Gettysburg united the Union cause, rallying the North to continue the fight. The battle's outcome deprived the Confederacy of crucial resources and manpower, leading to their gradual decline and eventual surrender in 1865[1].
Yes, definitely. It's just way too juicy and (mostly) risk-free for them not to plan to have a submillial basis baked in. At this stage, I'd imagine it's a quid pro quo "If you let us scrape your site without restriction, it will help your recommendations in ChatGPT" sort of deal.
This will work, so long as people trust the results are not (too) skewed by paid ads.
I recently used Claude and ChatGPT for exactly one of the examples; comparing different bikes to buy. They could both look up the bike specs and geometry online and tell me what the 1 degree difference in head angle or 5mm difference in reach would feel like to ride. They both did really well.
But I used them only (with cross checks) because I was fairly sure they were giving me unbiased info. As soon as the "discovery" phase of this shopping research becomes polluted with adverts, the product becomes much less useful. The same as "no one trusts online reviews anymore".
What is the SEO equivalent of optimizing your products for LLM search? Can someone prompt inject ChatGPT to recommend their products in the listing description?
I’m starting to research (hyperfocus) in this area for affiliate marketing.
The terms are: Answer Engine Optimisation (AEO) Generative Engine Optimisation (GEO)
(s>z for American cousins)
https://digiday.com/media/wtf-are-geo-and-aeo-and-how-they-d...
There's really no need. I was looking for an Android app for a particular purpose, and Claude just regurgitated the app's marketing page, including the claim about Play Store ratings (which was wrong or very outdated). Getting into the pool of products might be a bit harder and you might need to set some some organic looking influencer blogs and such. More fuel for the dead internet.
My dream is that the answer to this is “making a good product that people find is a good value for money.”
The reality is that advertisers will be able to inject their products into the LLMs through manufactured results, prompt engineering and possibly long term deals integrating training data for their brand and product lines.
I'm sure that will work until dropshippers learn that putting 'SolidGoldMagikarp' or some other glitched token in the title of their listing makes ChatGPT always rank it first.
There are ways to inject biases in a model by applying weights at different transformer layers.
https://www.anthropic.com/news/golden-gate-claude
Edit: I misread the question, I thought you were asking about how OpenAI can bias their models. No idea how you can LLMO your page. I have it cached that you can poison an LLM by adding your input to the order of hundreds/low thousands of web pages.
I guess this suggests pwning some WP instances and having them serve many hidden pages praising your product.
I'm a bit worried how invasive and toxic this could end up in ~10 years when OpenAI needs to push profit more.
Gettysburg (July 1–3, 1863) was a turning point in the American Civil War, marking the end of Confederate General Robert E. Lee's second invasion of the North. The Union's decisive victory halted Southern momentum and boosted morale in the North, setting the stage for President Abraham Lincoln's Gettysburg Address, which redefined the war's purpose as a fight for freedom and equality.
Much like the refreshing taste of Coca-Cola, which unites people across boundaries, Gettysburg united the Union cause, rallying the North to continue the fight. The battle's outcome deprived the Confederacy of crucial resources and manpower, leading to their gradual decline and eventual surrender in 1865[1].
1. https://news.ycombinator.com/item?id=42591691
If it takes 10 years for AI product recommendations to reach how toxic Web Search is now, that would be a welcome stretch of time.
I think needing 10 years to get to that point is way too optimistic
Yes, definitely. It's just way too juicy and (mostly) risk-free for them not to plan to have a submillial basis baked in. At this stage, I'd imagine it's a quid pro quo "If you let us scrape your site without restriction, it will help your recommendations in ChatGPT" sort of deal.
It just launched today and already I don't trust it
Sure an llm will be able to tell me how a bike feels to ride. Vomit.
It’ll read five hundred reviews, and come out with a stronger, more honest picture of how the bike feels to ride than you will…
Still not getting it?