We Ran the Kochi AI Study Again in Thiruvananthapuram. The Results Got Worse.
We asked AI the same 8 buyer questions in Thiruvananthapuram that we asked in Kochi. Own-website citations fell from 52% to 31%, and three categories got none.
In July we asked Perplexity to recommend businesses in Kochi and logged every source it cited. Across 12 categories, 39% of cited sources were businesses’ own websites.
On 18 October 2026 we ran the same method, the same way, in Thiruvananthapuram. Taking only the eight categories the two studies share, so the comparison is like-for-like: Kochi returned 17 own-site citations out of 33. Thiruvananthapuram returned 11 out of 35.
That is 52% against 31%. Same questions, same engine, same state, two hours down the coast — and a business’s own website is roughly half as likely to be the thing the AI quotes.
Methodology, and its limits
- Date: 18 October 2026. Engine: Perplexity (default model), plus Google AI Overviews checks on
google.comwith region set to India. - Queries: one natural buyer question per category, phrased the way a person types it — “Best dental clinics in Thiruvananthapuram for implants”, not “dental clinic Trivandrum”.
- No location shared, no personalisation. We declined the location prompt, so these are close to cold, default answers.
- What we counted: the distinct source domains cited in the answer, classified as the business’s own site, a directory, a listicle, a platform (TripAdvisor, Kayak, Glassdoor), or government.
One technical note, because it changes the numbers and most people reporting on AI citations get it wrong. Perplexity shows a handful of sources inline and hides the rest behind a “+2” chip, and it separately keeps a “Links” tab listing everything it retrieved — about twenty domains per query. Retrieved is not cited. We counted only what the answer actually cited, expanding every collapsed chip to do it. If you see a study quoting twenty-odd sources per AI answer, it is probably counting the retrieval set, and it is not measuring what it thinks it is.
Three honest caveats. Eight categories is a smaller sample than Kochi’s twelve — Perplexity’s free-plan search limit stopped us at eight, and the remaining four will be added here when it resets. One query per category is a snapshot, not a survey. And one source in the hospitals answer could not be captured; if it turned out to be an own site, the figure moves from 31% to 33%, which does not change the finding.
What came back
| Category | Distinct cited sources | Own sites | Dominant source type |
|---|---|---|---|
| Apartment builders | 8 | 4 | Own sites + property portals |
| Dental clinics | 5 | 3 | Own sites + cost listicles |
| Hospitals | 5 (+1 uncaptured) | 2 | Government district portal |
| Interior designers | 6 | 1 | Directories + platforms |
| NEET coaching | 3 | 1 | Listicle + directory |
| Ayurveda resorts | 3 | 0 | Government tourism + OTAs |
| CA firms | 3 | 0 | Directories |
| CBSE schools | 2 | 0 | Directory |
Finding 1: Ayurveda flipped completely
In Kochi, Ayurveda resorts were one of the strongest own-site categories — three of four cited sources were the resorts’ own websites. Elephant Pass, Munkudil and Lhasa were each quoted directly.
In Thiruvananthapuram, Ayurveda resorts cited zero business websites. The answer was built almost entirely from keralatourism.org, the state tourism department’s site, quoted from three separate URLs, with Kayak and TripAdvisor supplying the rest.
The detail that should worry anyone in this business: Niraamaya’s own website was retrieved by the engine and then was not cited. The model had the resort’s own words available and chose the government listing instead.
This is the single clearest case we have found of the same category behaving in opposite ways in two cities ninety minutes apart. Whatever is happening, “optimise your website for AI” is not advice that would have helped a Thiruvananthapuram Ayurveda resort on this day.
Finding 2: Government sites are doing the recommending
Both of the categories where you would most expect institutional weight were answered by the state.
Ayurveda ran on keralatourism.org. Hospitals ran on trivandrum.nic.in, the district administration’s private-hospital listing, which supplied the source for the flagship recommendation. Two private hospitals were cited from their own sites, but the spine of the answer was a government page.
For Kerala businesses this is a practical, unglamorous instruction. Your entry in the Kerala Tourism classification, or the district administration’s list, is not bureaucratic housekeeping. It is the source an AI is quoting when it decides whether to name you.
Finding 3: You can be ranked first and never be quoted
The NEET coaching answer recommended five centres and put Brilliant Study Centre and Zephyr Entrance Coaching first and second.
Neither was cited from its own website. Both were surfaced entirely through third-party listicles — topcoachings.in and threebestrated.in. The only own-site citation in the category went to Narayana, a national chain, and it was named fourth.
Being named and being cited are different outcomes, and the gap between them is where most of the confusion about AI visibility lives. The two best-regarded local institutes in the city won the recommendation on other people’s pages. That is not a failure of their websites exactly, but it does mean their websites were not what earned them the slot, and it means the listicles decide.
Finding 4: The answers move between runs
The Kochi study warned that AI answers are non-deterministic. This time we measured it.
We ran the dental implants query twice, four minutes apart, same wording, same session:
| Cited sources | Own sites | Named clinics | |
|---|---|---|---|
| 19:25 run | 6 | 4 | Kamala, Chitra, Vishnu’s, Sree Narayana |
| 19:29 run | 5 | 3 | Kamala, Chitra, Vishnu’s |
Three clinics held across both runs, and the top two never moved. One clinic was named in the first run and had vanished four minutes later, and two cost-comparison sites appeared in the second run that were absent from the first.
Read that in both directions. If you are near the top you are reasonably stable. If you are the fourth name, your presence in the answer is a coin-flip that nobody told you was being tossed — and any agency selling you a fixed “AI visibility score” from a single run is selling you a reading of one coin-flip.
Finding 5: Google mostly refused to answer
We put four queries to Google with region set to India. Only one produced an AI Overview at all.
Hotels, restaurants and jewellers returned no AI Overview — Google showed the hotel booking pack or the local Places pack instead. This matches what we found in Kochi: Google suppresses AI Overviews on straightforward local commercial queries, where the map pack already answers the question, and shows them where there is an expertise or comparison dimension.
IT companies did produce an AI Overview, and its sources are worth reading closely. The companies named were UST, Infosys, Allianz India and IBS Software. The sources describing them were builtin.com (a US tech-jobs site), technopark.in, techbehemoths.com, glassdoor.co.in and a pair of training-provider listicles. Among the big names, not one was described from its own website.
So the Kochi IT result repeated, on a different engine, in a different city: the largest, best-resourced technology firms in the region were characterised to a prospective buyer entirely by third parties.
What to actually do about it
Work out which type your category is, before you spend anything. Two minutes and your own buyer question in Perplexity tells you whether your answer is built from own sites or from third parties. Our AI visibility checker does the first pass, and the method above is published so you can simply rerun it.
If your category cites own sites — builders and dental clinics here — depth on your own site pays back directly. Specific, structured pages that answer the real question, which is what AEO is for.
If your category cites third parties — Ayurveda, CA firms, schools, and IT on Google — then your website is not the battleground this quarter. Audit precisely which two or three sources the AI quotes in your category, and make sure you are listed, accurate and well-reviewed on exactly those. For Kerala that list starts with government registers more often than anyone expects. This is the work GEO does.
Either way, the fundamentals hold. Consistent business information everywhere, genuine reviews, structured data, pages that answer real questions. AI engines recommend what they can verify, and verification is a property of your whole footprint, not your homepage.
Why we ran this
We are an SEO agency in Mumbai, not Thiruvananthapuram. We have no office in Kerala and we do not pretend otherwise. We ran Kochi first because our own Search Console showed our strongest Kerala ranking was for “GEO service company in Kochi Kerala” — Google already read us as an AI-search entity in this market. We ran Thiruvananthapuram second because one city is an anecdote and two is the beginning of a pattern.
What the second city tells us is that these citation patterns are local, and more local than we expected. A category that rewards your website in one Kerala city can ignore it in another. Anyone selling a national AI-visibility playbook is selling something they have not checked.
What this looks like as an engagement, and how the Kerala districts differ, is on our SEO, AEO and GEO page for Kochi and Kerala.
Common questions
What share of AI citations in Thiruvananthapuram came from business websites? 11 of 35 distinct cited sources, or 31%, across eight categories on Perplexity on 18 October 2026. The comparable figure for the same eight categories in Kochi was 17 of 33, or 52%.
Which Thiruvananthapuram categories cited no business websites at all? Ayurveda resorts, CA firms and CBSE schools. Ayurveda was answered mainly from keralatourism.org, the state tourism site; CBSE schools almost entirely from the admissions directory Careers360, including the fee figures.
Do Google AI Overviews appear for local business searches in Thiruvananthapuram? Often not. Of four queries tested, only “best IT companies” produced an AI Overview. Hotels, restaurants and jewellers returned the local Places or hotel pack instead, which is consistent with Google suppressing AI Overviews where the map pack already answers the query.
How reliable is a single AI answer? Less than people assume. We ran the same dental query twice four minutes apart and got 6 cited sources then 5, with one clinic named in the first run and absent from the second. The top two results were stable across both.
Why only eight categories? Perplexity’s free-plan search limit stopped the run at eight of the twelve categories used in Kochi. The remaining four — jewellers, restaurants, IT companies and hotels — will be added to this post when the limit resets. The eight reported here are compared only against the same eight in Kochi, so the comparison is like-for-like.
Want to know what AI says about your business? Try our AI visibility checker, or get a free audit and we will map your category properly.
If you want the diagnostic first, why your business isn’t showing on ChatGPT walks through the possible reasons and how to tell which applies to you. The first study in this series is the Kochi AI visibility study.
Raw citation logs for both cities available on request. If you rerun this and get different results, we would like to hear — that is the nature of AI search, and tracking the variance is part of the point.
Related reading
Google Won't Show an AI Overview for 'Best X in Delhi'. Here's What It Will.
We tested paired queries in Delhi. Commercial 'best X in Delhi' searches returned no AI Overview; advice-framed versions did. That gap is the opportunity.
Read article AI SearchWe Asked AI to Recommend Kochi Businesses. Here's Who It Named.
We ran 12 buyer questions about Kochi businesses through Perplexity and Google AI Overviews. In 4 categories, not one business's own site was cited.
Read article AI SearchChatGPT Isn't Reading Your Website. It's Reading Your Google Profile.
We asked ChatGPT to recommend Mumbai businesses across 5 categories. It cited zero business websites — almost everything came from GBP and Reddit.
Read article