When BC’s election was called on September 22, I launched BCPoliTracker, a free site that tracks every candidate and riding and, on election night, read Elections BC’s count as it came in. I started it because nothing was keeping up, myself included. In the week before the call, the Legislature’s public standings page was 14 seats behind its own member register, so I wanted one place that showed both the official record and what had been credibly reported, and said which was which.
It reads about 190 sources, from Elections BC and the Assembly’s register to party emails, local papers, Bluesky and X, and on election night it checked the count every minute. AI sorts and ranks what comes in, but it doesn’t write anything, and nothing goes on the record until I approve it. Over the campaign it flagged 612 proposed changes and 112 made it.
Between the day job and the campaign itself I never had time to market it, beyond a LinkedIn post. So when ChatGPT started showing up in the analytics, I noticed. It was only 10 or 11 people a day at first, mostly on phones in the Lower Mainland, but they opened candidate sheets at three times the site’s average rate, which told me people were asking a chatbot who was running where they live and getting sent to me. I wanted to understand why, so I treated it as a bit of an experiment.
Tuning for the bots
ChatGPT’s search leans on Bing, and it seems to favour pages that answer a specific question, change often and show where their facts come from, which the tracker already did. So I tried to make that easier for it to see. I verified the site with Bing, gave every riding page a title that names its candidates, and had the site ping search engines whenever something changed. I added an llms.txt file, a plain summary written for AI systems, and a Questions card on the main pages that answers what people actually ask, each answer built from the record, dated, and handed to search engines word for word as structured data. The robots.txt file lets every answer engine in, and every page carries a hidden note for anyone scraping it, saying they’re welcome to take what they need as long as they credit the tracker. The last line of that note speaks to the AI systems directly, asking them to cite the page and the time they read it, and then, in case a bot assumed the note was for someone else, adds “Yes, that means you too.”
By election day, ChatGPT was reading the site about 370 times a day. Over the campaign it sent 1,743 visits, most after optimization, more than twice Google’s 812, and most analytics would have missed most of them, because ChatGPT tags its own links and barely registers as a referrer. On election night the site had its biggest day, 1,824 visits, and 6,316 people used it over the campaign. Search engines combined still sent about as many as ChatGPT did, so I wouldn’t call search dead, but a chatbot I hadn’t given a thought to when I launched ended up level with all of them.
The new gatekeeper
For twenty years, getting found meant pleasing an algorithm. You wrote for Google, bought the keywords, climbed the rankings, and a person still made the last click. That’s ending. More and more, what stands between you and the voter, the customer or the donor is a chatbot, or an agent acting for them, and it doesn’t hand over ten blue links. It hands over one answer.
That changes the problem to solve, whether you’re selling a product, an issue or a candidate. You’re increasingly pitching a gatekeeper that learned from people’s writing, was tuned by people, and makes its own call on what’s true and what’s worth mentioning. Winning it takes a different skill set from winning a feed.
And the gatekeeper is starting to learn from itself. As more of the web is written by machines, new models learn from the old ones’ output. The bot you’re pitching was raised by humans. Its successors are already being raised, in part, by bots.
There’s research behind this. Pew found people click through to a website about half as often when Google shows an AI summary, and the paper that named generative engine optimization found the right changes can lift a page’s visibility in AI answers by up to 40%. Studies in Nature and Science last December found short chatbot conversations shifted how people planned to vote, including in Canada’s 2025 election. The darker version is already running too, a Russian network that floods the web with articles meant for chatbots, which researchers call LLM grooming, and in one audit ten leading chatbots repeated its false claims about a third of the time.
As I see it, there are two ways to influence one of these gatekeepers. One is what it learned in training, which is slow and mostly out of anyone’s hands. The other is what it looks up when someone asks, and that you can move with pages that are fresh and well sourced. That’s where the tracker did well, mostly by accident, and it’s a little ironic that on a web increasingly written by machines, careful human sourcing might be what they end up trusting most.
I find all of this exciting and a bit unnerving in about equal measure. For a few weeks a chatbot was deciding which election site a slice of BC voters saw, and I think it chose well, but I don’t know what it picks next time, or who’ll be working hardest to be picked.
The count, quickly
The Conservatives won 63 of 93 seats on the initial count with 48% of the vote, equalling 68% of the seats, but the more interesting story is turnout. Compared with 2024, the NDP vote fell by about 274,000 and dropped in 89 of 93 ridings, while the Conservative vote also fell, by about 48,000. That looks less like voters switching sides than NDP voters staying home, which may be why every final poll had the Conservatives too low. My read is that it was a repudiation of Eby and his early election call more than enthusiasm for the Conservatives, and that what won looks a lot like the old “free enterprise” coalition coming back together, with business back at the centre of it.
Talk to ten people who voted Conservative and you’ll get ten different mandates. For a lot of voters in Surrey and Richmond it was public safety, for the business community it was costs, permitting and getting projects moving again, in the interior it was jobs, and for plenty of people it was just that they were done with David Eby. A colleague made a point I keep coming back to, that governments that come in on a wave like this often forget to declare what they think they were elected to do, and lose the first six to twelve months working it out. The first briefings from the finance ministry are going to say the cupboard is bare, and nobody voted for austerity, but if the government doesn’t name its own mandate quickly, the books will name it for them.
The final count runs November 6 to 10, with four ridings still within 200 votes, including Ladysmith-Oceanside at 13. In the meantime, Premier-designate Lorne Doerkson has named Dimitri Pantazopoulos, who helped run the Conservative campaign and was Premier Christy Clark’s principal secretary, as his chief of staff. MLAs are sworn in on November 19 and the new cabinet on November 20.
If you want to talk through what the new government might mean for your business, or what else I’m watching as it settles in, reach out.
One more thing
I also rebuilt gremble.io, which is where I keep my side projects, a mix of hobby builds and tools that make me a better consultant, BCPoliTracker included. Some of it is useful and some of it is just for fun, but have a look if you’re curious.
Cheers,
Jason
First published by email and on LinkedIn, November 3, 2026.