A procurement manager in Durban needs a bonded warehouse. A few years ago she would have searched Google, scanned the first page and made three phone calls. Today there is a reasonable chance she types the question into ChatGPT instead, reads a paragraph of considered-sounding advice, and works from the handful of companies it names.
That shift is quietly rewriting how supply chain businesses get found. And it comes with a complication most operators have not yet noticed: the answer she gets depends heavily on which tool she opens. Ask ChatGPT, Google’s AI Overviews and Perplexity the same sourcing question and you will often get three different sets of cited companies, sometimes with barely any overlap.
This is not one of the systems getting it wrong. As a recent analysis by Johannesburg digital agency IMS sets out, each platform runs a genuinely different process for deciding which sources are trustworthy enough to name. Understanding those differences matters more to a freight forwarder or 3PL than knowing the acronym for it.
ChatGPT: Being Read Is Not the Same as Being Named
When ChatGPT searches the web, it reads far more than it credits. Independent analysis of its browsing behaviour has found that only a fraction of the pages it retrieves end up cited in the final answer. The rest inform the response without ever being named. For a logistics company, that is the difference between a shipper seeing your name and a shipper seeing a competitor’s while your website quietly supplied the background.
ChatGPT also does not search the live web on every question. It is far more likely to go looking when the query carries commercial or comparative intent; “best cold chain provider,” “Transnet vs road freight costs,” “cheapest customs clearance Cape Town 2026”, than when someone asks it to explain what a bill of lading is. Which is convenient, because commercial intent is exactly where supply chain buyers sit.
The pages that survive the cut tend to share the same traits: they answer the specific question directly, they carry concrete figures rather than general claims about service excellence, and they are structured so a passage can be lifted out cleanly.
Google AI Overviews: A Page One Ranking Is Not Enough
AI Overviews sits on top of Google’s existing search index, but qualifying for the AI panel is a separate process from ranking. A logistics site can hold position two for a competitive term and still be skipped, while a page sitting at position eight gets quoted because it states the answer in two clean sentences with a number attached.
Researchers studying AI Overview outputs describe the process as a funnel: a large pool of candidate pages narrowed through semantic relevance matching, then authority and expertise filtering, then a final re-ranking before a small number of sources are stitched into the summary. Depth counts here in a way it does not for ChatGPT. Google appears to favour sites showing sustained expertise across several interlinked pages on a subject, rather than one strong article standing alone.
For an operator, that has a practical translation. A single well-written page about your temperature-controlled fleet is worth less than a cluster of connected pages covering cold chain compliance, load monitoring, last-mile handover and the regulations that govern them.
Perplexity: Built to Cite From the Ground Up
Perplexity is the outlier, usefully so. Citation is not a feature added to a chat product; it is the product. It runs a retrieval-augmented pipeline with multiple ranking layers scoring relevance, freshness, factual accuracy and structural clarity before anything reaches the answer.
It is also the most selective of the three. Perplexity will typically visit around ten pages for a query and cite only three or four. Structural trust signals carry real weight: named authors, visible editorial standards, and claims corroborated across more than one independent source rather than appearing on a single company page.
This is where trade publications, industry associations and conference coverage become genuinely valuable to a logistics business, not as vanity placements, but as the independent corroboration these systems are specifically looking for.
Why the Three Disagree
Put the pipelines side by side and the disagreement stops being mysterious. ChatGPT is deciding whether a page is worth naming after it has already read it. Google is applying an authority-and-extractability filter on top of an index built for a different purpose. Perplexity is built around sourcing and rewards signals the other two barely weigh.
The divergence is not random, either. A 2024 audit of ChatGPT, Bing Chat and Perplexity by researchers Alice Li and Luanne Sinnamon, published in the Proceedings of the Association for Information Science and Technology, found that generative search systems lean heavily on news, media and business publications for their sources, and showed measurable commercial and geographic bias in which sources they use to support claims.
Geographic bias deserves attention in this market. A system weighted toward North American and European business media is a system that may not surface the South African freight forwarder that is genuinely the right answer to a South African question.
What the Research Says Actually Works
The most rigorous evidence available comes from “GEO: Generative Engine Optimization,” a study by Pranjal Aggarwal, Vishvak Murahari and colleagues at Princeton, Georgia Tech and IIT Delhi, presented at KDD in 2024. The researchers built a benchmark of roughly 10,000 real queries and tested nine content optimisation strategies against generative engines.
The strongest performers were not keyword tactics. Adding statistics and adding direct quotations were among the most effective changes tested, improving visibility by roughly 30 to 40 percent against an unoptimised baseline. Authoritative language and explicit sourcing also helped. Keyword stuffing did close to nothing.
That finding sits comfortably with how the logistics sector already communicates. On-time delivery percentages, claims ratios, dwell times, fleet utilisation figures, tonnage handled, accreditation numbers; the industry is unusually rich in exactly the kind of concrete, quotable detail these systems reward. Most of it never makes it onto a company website, where it is replaced by phrases like “world-class logistics solutions.”
What This Means for the Industry
Three consequences follow.
The first is that vague marketing language is now actively costly. A page claiming end-to-end excellence gives an AI engine nothing to quote. A page stating that a facility holds 12,000 pallet positions, operates to a stated temperature tolerance and cleared a specific volume of customs entries last year gives it something to work with.
The second is that being visible on one platform tells you nothing about the others. Because the pipelines differ, and because independent research confirms real bias in what they select, a company cited confidently by Perplexity may be invisible in ChatGPT and Google’s AI Overviews. Each has to be checked on its own terms.
The third applies to the buying side. Supply chain managers using these tools to shortlist providers should treat the results as a starting point shaped by a particular set of preferences, not a market survey. A carrier’s absence from an AI answer says something about its web presence. It says very little about its trucks.
The industry has spent a decade learning to be found on Google. The engines that increasingly sit between a shipper and a supplier now work differently from each other, and differently from search. That is worth understanding before the next tender or RFP goes out.