It is 7:14pm on a Friday. You and three friends have thirty minutes before the first cocktail. Someone opens a generic AI chatbot, types "best bar near Soho", and gets a list of five places — two of which closed during the pandemic, one in the wrong city, and the other two ranked alphabetically. The FunSpot concierge exists because that experience is broken.
FunSpot is an AI concierge specialised in going out. It is grounded on real venues, real reviews, and real press — not a stale training snapshot. This piece explains how the concierge works, what it will and will not do, and why "grounded" is the whole ball game.
Why Generic AI Chatbots Get Nightlife Wrong
Large language models are brilliant at language and terrible at freshness. Ask ChatGPT for a rooftop bar in Lisbon and it will confidently describe a place that reopened under a different name two years ago — or has never existed at all. This failure mode, called hallucination, is a structural limit of a model that cannot look at the world in real time.
Going out is the worst possible domain for hallucination. Venues close, rebrand, change neighbourhoods, lose licences, start doing karaoke on Tuesdays. A stale answer does not just embarrass the chatbot — it wastes your evening.
What Makes FunSpot's Concierge Different
Grounded on Real Venues, Real Press, Real Reviews
Every recommendation FunSpot returns is tied to a specific venue in a database curated from structured sources: official venue pages, press coverage from outlets our editorial team trusts, and verified user reviews. The concierge can only recommend places it can cite. If it cannot cite anything, it says so and asks a clarifying question instead of inventing.
The language model in FunSpot is a reasoning layer on top of the database. It reads your query, ranks candidate venues, and then stitches together a reply in the voice of a friend who knows the city. But it never returns a venue that does not exist in the index.
No Signup Wall to Browse
You can use the concierge without creating an account. That is a deliberate choice from our beta terms. Social features — matching plans with friends, saving lists, leaving reviews — require a verified account, but the core "where should we go tonight" experience does not.
Try these queries in the concierge:
- Romantic dinner for an anniversary in Edinburgh
- Late-night cocktail bar with rooftop seating in Manchester
- Quiet coffee spot for a first date in Bristol
- Queer-friendly venues in Glasgow this weekend
Ask It Anything in Plain Language
The concierge takes natural-language queries — the longer and more specific, the better. "A bar" returns less useful answers than "a bar under £10 for a pint, with outdoor seating, close to King's Cross, open past midnight on a Wednesday". Constraints compound. The model uses each one to narrow the pool of candidates before ranking.
When you ask the concierge something it cannot answer — a city we have not indexed yet, a niche no venue in our database fits — it tells you, instead of guessing. That sounds obvious. It is not how generic chatbots behave.
Ranked by Semantic Match, Trust Score, and Recent Activity
Once candidate venues are pulled from the database, three signals shape the ranking:
- Semantic match — how closely each venue matches the intent of your query, not just the keywords. A bar that describes itself as "lively" is a different result than a bar described as "quiet" even if both have "cocktail" in the name.
- Trust Score — an editorial signal combining verified reviews, press mentions, and operator claim status. Venues with a claimed profile and recent reviews rank higher than unclaimed listings with only historical data.
- Recent activity — check-ins, recent reviews, and live event schedules are weighted heavier than year-old data. A cafe that closed six months ago sinks even if it has great historical press.
For Business Owners: Claim, Verify, Promote
Venue operators interact with FunSpot on a separate surface — app.mw3.biz/funspot — where they can claim their listing, verify ownership, respond to reviews, and run promotional vouchers. The business surface is deliberately not part of the consumer site. That separation keeps the concierge honest: operators cannot pay for placement in the ranking.
The concierge is only as good as the index behind it. A hallucinating recommendation is worse than no recommendation at all — it destroys trust, and trust is the only currency a going-out site has.
That is why we invested in the index first and the AI second. Plenty of startups have the order reversed.
Open the concierge and try a query. No signup required.
Run a venue? Claim it on app.mw3.biz/funspot.