We’ve been here before: shiny new tech pulling people away from tools they trust
Today’s generative AI tools — ChatGPT, Claude, Gemini — remind me of Google Maps between 2008 and 2012. I don’t just mean in terms of hype or innovation. I mean in how quickly people are abandoning their trusted tools in favor of something shiny and new… even if that new thing isn’t fully reliable yet.
I had a front-row seat to this phenomenon once before. Back then, I worked at Google Maps — first as a Product Specialist, later as a UX Researcher. I saw a technology go from niche to daily essential, reshaping habits overnight. I also saw what happens when a platform scales faster than it can safeguard accuracy: trust gets chipped away.
That’s where generative AI is right now. Millions of people are swapping their tried-and-true single-use apps, search engines, and programs for AI tools. And just like Maps, these platforms are built on massive “clusters” of data stitched together from countless sources. In both cases, what matters most is which piece of data “wins” in the cluster hierarchy — the answer the user actually sees.
When the wrong truth wins? Users get lost. Sometimes figuratively, sometimes literally. And once that trust is gone, it’s an uphill climb to earn it back.
⸻
When AI Trips Over the Basics
Even tiny errors can trigger massive doubt.
A few weeks ago, I was applying for small business grants. One application required my answer to be under 3,000 characters. Easy enough. I dropped my draft into Claude.ai and asked it to shorten it.
Claude gave me a neat, concise version. I pasted it into the form — error message. Still too long.
I checked in Google Docs: nearly 4,000 characters.
I went back to Claude. Tried again. Still too many characters.
Only after a third attempt — where I gave Claude the Google Docs character count — did I finally get a version that worked.
It was such a simple, objective request: count characters. And I blindly trusted Claude to get it right. That tiny failure triggered a bigger question: If AI can be wrong about something this basic, what else am I letting it handle without double-checking?
In Google Maps terms, this was like giving me directions to a restaurant… that doesn’t exist. The “wrong truth” had won the cluster. And if that wrong truth isn’t corrected quickly, it stays in circulation, misleading the next person who comes along.
⸻
The Evolution of Getting Lost
From AAA fold-outs to MapQuest to glitchy early Google Maps.
In the ’90s, I navigated California with AAA fold-out maps. In 2002, I printed MapQuest directions to drive from my home in San Rafael down to UCLA.
By 2008, the world was shifting to Google Maps. It was exciting, but it wasn’t perfect. And for the first time, I saw what happens when a platform becomes essential before it becomes truly reliable.
⸻
Google Maps’ Growing Pains
Bad info, spam, and missing data were everyday problems.
In 2010, we knew that about 40% of all Google searches had local intent (source). People were looking for restaurants, hotels, plumbers, locksmiths — not just street addresses. Google Maps was evolving into a full-blown local search engine.
The rapid growth was outpacing its readiness.
Our team (shoutout to ConOps) faced the same three top complaints over and over:
- Incorrect information
- Spam
- Missing data
The underlying cause was always the same: Maps pulled from multiple data sources, then clustered them into a single record. The cluster hierarchy determined which data point appeared as the “truth.” Sometimes it nailed it. Sometimes it was completely wrong.
I presume AI works the same way today. A “winner” delivered with total confidence. And if it’s wrong, you may never know. Other times, AI does the opposite and dumps everything it can find in your lap, whether or not it’s valid.
Case in point: I once asked ChatGPT for current promo codes at a store. It confidently handed me 14 codes. Not a single one worked.
⸻
Outdated Info: My Trip to a Nonexistent Costco
When the “truth” is wrong, the whole journey falls apart.
I once printed Google Maps directions to a Costco in Southern California that turns out didn’t exist. I ended up lost in an unfamiliar part of Los Angeles, digging through my glove box for an old AAA map, to end up driving nearly an hour back to where I started.
I see strong parallels to AI:
- Outdated info
- Wrong sources winning the cluster
- Users assuming the “truth” they’re given is correct
In Google Maps’ case, outdated info meant closed businesses, wrong phone numbers, incorrect hours, or moved locations still showing up at old addresses. Sometimes the errors came from bugs. Sometimes from businesses being merged incorrectly. And sometimes from user-generated reviews showing up under the wrong listing — harmful for the business and confusing for the customer.
When brick-and-mortar business owners spotted inaccurate data, they escalated — fast. The most vocal made their frustration public, sometimes through the press, which put even more pressure on the Google Maps team to act.
In response, we streamlined both how we collected reports and how we prioritized and resolved them internally. We rolled out several initiatives to tackle inaccuracies head-on:
- User-facing feedback tools like Report a Problem
- Direct engagement with owners via Google My Business
- Establishing a reliable “ground truth” through Street View imagery and dedicated verification teams
⸻
Spam: Funny Until It’s Not
The slow erosion of credibility.
Spam in Maps was sometimes funny — like “Sexy Escort Services” listed at 1600 Pennsylvania Avenue — but also serious. Fake listings and keyword stuffing eroded trust in the platform for users. Worse, they actively hurt legitimate business owners by pushing them lower in search rankings, costing them visibility, customers, and revenue.
The hard questions we faced then are the same ones AI will face soon:
- How do you handle malicious actors?
- How do you verify legitimacy at scale?
- How do you surface legitimate but less visible sources?
Over time, Google Maps rebuilt credibility through a careful balance of inputs: authoritative sources, real-world signals from Street View and geolocation data, and user contributions — edits, error reports, reviews, and behavioral patterns — all layered with verification.
⸻
Missing Data: The Silent Trust Killer
What you can’t find is as damaging as what’s wrong.
Sometimes the issue wasn’t wrong data — it was no data at all. In Google Maps, that meant not finding my brand-new home’s address in the system. In AI, it’s when the model simply doesn’t know an answer but still tries to give you one.
Maps caught up through Street View, user-submitted corrections, and business-owner updates. Back then the process was slow — we were building the plane while flying it. A decade ago, the process was slow. But in 2025, slow is unacceptable. After stepping away in 2024, I came back to what feels like a completely new internet than I left it — things are moving at a pace that makes a decade ago look glacial. AI will need its own equivalent of “ground truth” to close this same gap.
⸻
How Google Maps Rebuilt Trust and closed the gap
Accuracy, feedback loops, and visible responsiveness.
Google Maps recovered by:
- Sourcing information from those closest to it (owners, locals, authorities)
- Building robust, visible feedback systems with a closed loop
- Investing in verification across multiple sources
It wasn’t fast, but it worked — because it made the system feel responsive and accountable.
⸻
AI Needs Its Own “Report a Problem”
Trust can’t survive without feedback and transparency.
Right now, AI tools have minimal feedback loops, and even less transparency. Where’s the “Report a Problem” button for a wrong answer in ChatGPT or Claude? Do I just tell the machine that information is wrong like I did with the character count for Claude? I get the sense it goes into a black hole — am I right? Where’s the follow-up that shows your feedback was acted on?
Without these, AI risks the same fate as MapQuest: revolutionary for a time, but ultimately replaced by something more reliable.
The winners in AI won’t just be the fastest — they’ll be the most trusted. And trust is built on accuracy, recency, and a visible willingness to correct mistakes.
⸻
The Road Ahead
The next AI winners will win on trust, not speed.
After my Claude experience, will I trust it for critical tasks again? Probably not. That’s the cost of even small, preventable errors.
If AI companies want to lead the next decade, they’ll have to:
- Make sure the right information wins the cluster hierarchy.
- Give humans a real say in defining that “truth.”
- Build transparent, responsive feedback systems.
Otherwise, we’ll all be taking a lot more metaphorical detours to non-existent Costcos.
⸻
Author’s Note
I’ve simplified many factors here. Speed, reliability, and UX matter enormously. Google’s success came from guiding users seamlessly from search → business → directions → arrival, even solving adjacent problems like “where did I park?”
AI tools will need the same holistic thinking — functional utility that goes beyond query → response, much like Google Maps has done with features like ‘save parking spot’. This is where I see endless opportunity for platforms to build upon trust and accuracy and truly take AI to the next level.
The future leaders won’t just answer questions; they’ll anticipate needs, integrate into full journeys, and treat trust like a precious, fragile resource.
⸻
You don’t have to navigate this ambiguity alone! Partner with us to develop future-thinking solutions that will help you stay one step ahead of the competition.