Himabindu Lakkaraju

Himabindu Lakkaraju is on the faculty at Harvard University. With Aounon Kumar, she showed in "Manipulating Large Language Models to Increase Product Visibility" that a crafted strategic text sequence added to a product page could push a language model to name that product its top recommendation.

The authors tied the risk directly to fair market competition. If a hidden or adversarial pattern can swing a recommendation, then generative shopping and vendor advice inherit a new attack surface. That framing moved the paper out of a narrow security niche and into marketing and policy conversations. The full paper is the primary source.

For GEO practitioners, the study was a warning label. Techniques that look like clever optimization from one angle look like market distortion from another, especially when the user never sees the injected sequence.

The competition angle is what made the paper hard to ignore outside academia. A manipulated recommendation is not only a security curiosity. It is a market fairness problem for any category where buyers ask AI systems which product to choose.

The study became a reference point for the boundary the GEO field still debates. Legitimate work tries to make accurate, useful brand evidence easy to retrieve. Manipulation tries to coerce a model regardless of merit.

Platforms, agencies and researchers cite the paper when they need a concrete example of why spam and injection defenses matter in AI search, not only in classic web results. Lily Ray and other arbiters often sit on the same side of that argument when they flag prompt-injection patterns in the wild.

Lakkaraju's place among the architects is earned by defining a risk clearly enough that the commercial field cannot pretend it does not exist.

The paper also gave platforms a research citation when tightening defenses against recommendation-steering text. Academic clarity sometimes travels faster into policy than a marketing blog post does.

Start with "Manipulating Large Language Models to Increase Product Visibility" (2024). Read it as a warning paper, then compare its techniques with the tactics vendors are selling. The gap between those two lists is where ethics reviews should start.

The field still argues over where persuasion ends and manipulation begins. Lakkaraju and Kumar gave that argument a concrete experimental anchor instead of leaving it as a vibes debate on social media.