By Shravan Prasad

Welcome back to The Neurals. After investigating dynamic pricing in my last post, I couldn’t shake a nagging follow-up question: if the price I see is different from the price you see, is the product I see also different from the product you see? I decided to find out. What I uncovered about AI-powered search and product ranking is equal parts impressive, unsettling, and deeply revealing about who e-commerce is really optimized for.

The Search Bar Isn’t Neutral

Let’s start with a simple test. I searched for “wireless headphones” on three different e-commerce platforms — logged in, logged out, and from a fresh browser on a different device. The results were meaningfully different every time. Same search term. Different products at the top. Different prices featured. Different brands given prime real estate.

That’s not a glitch. That’s the algorithm working exactly as intended.

Modern e-commerce search is no longer a simple keyword-matching exercise. It’s a real-time decision engine powered by machine learning models that weigh dozens — sometimes hundreds — of signals simultaneously. And the signals it prioritises say a lot about whose interests the algorithm is actually serving.

What Signals Does the Algorithm Actually Use?

From my research into patent filings, published ML papers from Amazon, Zalando, and Alibaba, and my own experience working in e-commerce operations, here’s what’s likely going into the ranking model when you hit search:

Your personal history. What you’ve searched for before, what you’ve clicked, what you’ve bought, how long you lingered on a product page — all of it gets fed back into the model. If you’ve bought Nike three times, Nike likely ranks higher for you on generic terms like “running shoes.”

Conversion probability. The algorithm is trying to predict which product you specifically are most likely to buy. Not the best product. Not the highest-rated product. The one you’ll click and purchase. There’s a subtle but important difference.

Margin and profitability. This one doesn’t get talked about enough. Platforms give preferential ranking to products that are more profitable for them — either through higher commission rates, first-party ownership, or fulfilment economics. A product that earns the platform more per sale has a structural advantage, independent of quality.

Sponsored placement. Paid search results are baked into nearly every major platform, and the lines between organic and paid results have become deliberately blurry. On Amazon, multiple rows of sponsored results now appear before a single organic result. On Google Shopping, virtually everything above the fold is paid.

Inventory levels and logistics. Products that are well-stocked in nearby warehouses get a boost. Products with Prime eligibility or fast delivery promises rank higher — partly because of genuine customer preference for speed, and partly because it benefits the platform’s logistics economics.

Ratings and reviews — but not the way you think. Review scores matter, but not in isolation. A product with 4.3 stars and 10,000 reviews will typically outrank one with 4.8 stars and 80 reviews. Volume of social proof is weighted heavily, which creates a compounding advantage for established sellers and makes it genuinely hard for new entrants to break through.

I Tested It Myself

Inspired by my earlier investigation into my digital shopping profile, I ran a structured experiment. I created two browser profiles — one that had browsed extensively in the premium/luxury segment, one that had only engaged with budget products. I used both to search for identical terms across three platforms.

The results were stark. The “premium” profile was consistently shown higher-priced options first, even when lower-priced alternatives had better ratings. The “budget” profile saw more sponsored results from value brands and, interestingly, more aggressive discount messaging — “was ₹2,999, now ₹999” type framing — as if the algorithm had already decided this user needed a price incentive to convert.

Same search. Different user profile. Completely different commercial reality.

The Invisible Hand of Category Merchandising

Having worked as a Site Merchandiser at a large retailer, I know firsthand that there’s a human layer behind the algorithm that most shoppers never see. Merchandising teams manually boost certain products, create curated “featured” carousels, and set category-level rules that override pure algorithmic output.

Vendor negotiations play a significant role here. Brands that commit to higher marketing spend, exclusivity agreements, or better wholesale margins often receive visibility guarantees in return — a specific number of homepage impressions, category banner placements, or search ranking floors. It’s a commercial arrangement dressed up as organic relevance.

AI is now automating much of this. Machine learning models are being trained not just on what customers want, but on what commercial agreements and margin targets require. The algorithm becomes a mechanism for executing business strategy — with the customer’s apparent satisfaction as a constraint, not the primary objective.

So Is It Actually Bad for Consumers?

Here’s where I try to be honest rather than just alarming. The reality is nuanced.

Personalisation genuinely helps in many cases. If I’ve bought photography gear before and I search “bag,” seeing camera bags first is useful, not manipulative. Relevance is real value. The problem isn’t personalisation itself — it’s when the algorithm’s commercial objectives diverge from the user’s actual interests without transparency.

The more serious concern is what economists call “search neutrality” — the idea that a search result should reflect the best answer to your query, not the most commercially convenient one. In e-commerce, this principle is almost entirely absent. There is no obligation for platforms to show you the best product for your needs. They’re showing you the best product for their business.

When I searched “protein powder” on a platform where the retailer sells its own private-label brand, that private-label product appeared first — above products with significantly higher ratings from established brands. Is that the algorithm serving me, or serving the retailer’s margin? The answer, I think, is obvious.

What Can You Actually Do About It?

A few things I’ve started doing that I think are worth sharing:

Sort by average customer rating, not relevance. “Relevance” on most platforms is just a commercial ranking in disguise. Sorting by rating at least gives you a signal that’s harder to game — though not impossible, given review manipulation.

Scroll past the first page. The products on page two and three often represent genuinely good options that just lack the marketing budget or review volume to compete with established players. I’ve found some of my best purchases buried on page three.

Use incognito for price and product comparison. A clean session strips away your behavioural profile and often gives you a more neutral baseline to compare against what your logged-in session is showing you.

Be sceptical of “bestseller” and “recommended” tags. These labels are algorithmically assigned and often reflect sales volume or commercial priority, not product quality. They’re marketing language dressed up as editorial judgement.

The Bigger Picture

Looking across the posts I’ve written for The Neurals, a pattern is becoming hard to ignore. AI in e-commerce is extraordinarily sophisticated at one thing: converting your attention into a transaction. It knows what you’ve bought, what you’ll pay, and now, what to show you first to maximise the probability of a sale.

That’s impressive engineering. But it’s worth asking: who exactly is this optimised for?

The search bar felt like a tool you controlled. The algorithm has quietly turned it into a funnel it controls. Most of us never noticed the switch.

Next time, I’m going to investigate something I’ve been curious about for a while — how AI is reshaping returns and refund decisions, and whether the “hassle-free returns” promise is as neutral as it sounds. Spoiler: it isn’t.

If you found this useful, share it with someone who shops online. That’s everyone. See you next time.

— Shravan