We pointed Obsero at the same kind of questions a swing voter might ask in the run-up to last week's UK local elections. Who should I vote for. Who's winning. Who will win the next election. We ran those prompts and variants of them across ChatGPT, Claude, Gemini, Google AI Mode and Perplexity over the last four months, and pulled the answers and the citations behind them.

The headline finding: for every time Labour was named first in an answer, Reform was named first 11 times more.

Claude was the most likely to surface Reform first. Across the rest of the platforms the pattern held, just less aggressively. If you were undecided, and you turned to AI to help you make up your mind, the first party you saw was Reform.

That broadly matches the polls. Reform was always going to take seats. Labour and the Conservatives were always going to suffer. But AI didn't just match the polls. AI over-mentioned Labour and the Conservatives by three to six points compared to the polling data. The live web pulls in the current picture, but the underlying model still has a lot of ink spilled on the two traditional parties from years of historical coverage. There could also be guardrails in place to stop the platforms telling users who to vote for.

The size of the audience for this isn't hypothetical. A new LSE survey of 2,137 UK adults, published this week, found 20% of voters now use AI chatbots to find political information, more than double the figure from 2024.

Nearly 1 in 5 of those who asked AI about a politician said the answer changed their view.

AI is no longer at the edges of how voters research politics. It's part of the mainstream.

The sources behind those answers are the more interesting story.

Pollsters were cited in 27% of every answer. Wikipedia was the single biggest source overall. The BBC and The Guardian both made the top 10, and so did YouTube and Reddit at number nine and number five.

The mix changes by platform.

ChatGPT leaned heavily on news outlets and pollster reports, and barely touched Wikipedia. Perplexity and Gemini did the opposite, leading with Wikipedia. Claude reached for Wikipedia, Electoral Calculus and Statista. Pollsters were pulled in most by Claude and Gemini.

There's a wider point in here. Once you know two left-leaning publications are inside the top 10 sources for a politically charged prompt, you can guess what comes next. It's the same playbook we saw with social media and Cambridge Analytica. Politicians will start accusing AI platforms of bias and censorship. Some already are. Whether the accusations are fair or not, the platforms are going to have to defend themselves.

One small thing worth flagging. AI Overviews didn't return for any of these political prompts. Given how many years tech firms have spent in the firing line over election misinformation, I suspect that's a deliberate guardrail rather than a quirk of the data.

Sentiment in the AI answers tracked the political weather pretty closely. Claude was the most negative on Labour. Perplexity was the most positive. ChatGPT sat somewhere in the middle.

Some of the actual answers we pulled out:

These aren't tabloid newspaper headlines. These are answers being served up to voters who asked a question.

The bigger surprise was issue ownership. Even when the prompt didn't ask about specific issues, the AI answers brought them up anyway: cost of living, immigration, NHS, climate, education. We then looked at which party each platform credited as the owner of each issue.

Reform owned immigration outright. They came second on tax. The Conservatives owned the NHS at 59% and owned tax. Climate and energy were a near-tie between Reform and the Greens.

Labour didn't own a single issue. Not one. After 14 years out of power and a 2024 win that already feels like a long time ago, Labour are sitting in government with no issue they can point AI to as theirs. That's a problem you can't fix with a press release or a listicle.

Cost of living is interesting too. Gemini over-indexed on it at 34%. Claude was very low. The Google-owned platforms (Gemini and AI Mode) leaned much harder into cost of living than the others.

ChatGPT ads aren't mature enough to swing the next UK general election. The product is too new, the targeting too blunt, the inventory too light. By 2029 they will be.

Parties will be looking for ways to influence the answer, not just the search result, and to surface for the swing voter who's asked AI instead of reading a leaflet on the doormat. The advertising playbook is about to get a new chapter.

If you're a brand, look at this through that lens. Who owns your category in AI answers. Which sources are being pulled in. Which sentiment is showing up. The mechanics that just played out in a UK local election are the same mechanics that play out for any contested space. Including yours.

See the full report here.

OpenAI confirmed on 7 May that it's expanding the ChatGPT ad pilot into the UK, Japan, South Korea, Brazil and Mexico in the coming weeks. The pilot was previously US, Canada, Australia and New Zealand only. In the same week, OpenAI announced the rollout of a self-serve ads manager for US advertisers only.

This is significant for clients I work with on the GEO side. The real estate inside a ChatGPT answer just got more competitive and easier to buy into.

OpenAI also briefed press on what's next: third-party measurement, conversion tracking pixels (now live), cost-per-click bidding and cost-per-action bidding (in the works).

They've dropped the $50,000 minimum spend in the US to attract smaller advertisers. Asad Awan, the company's ads and monetisation lead, set out the roadmap.

The PR line is "this won't affect the experience, privacy is still strong". We've heard that line from every other platform that ever introduced ads, so I'd take it with a pinch of salt. The shape of the answer will change. The question is how fast.

Included within Google's "five new ways to explore the web with generative AI" announcement this week was something I think matters more than people realise: expert advice quotes inside Search and AI Mode.

Google will pull first-hand quotes from named sources directly into the answer, so a searcher can see what people who've actually been there think.

The barrier to creating content is non-existent. You can spin up commodity content at scale in an afternoon. What you can't fake at scale is experience. Knowing something about a particular subject, or being a creator who can produce a moment that inspires people, is going to be worth a lot more than the market currently thinks.

It's not just about how long you've been around. It's about whether you've actually done the thing you're writing about. If you have, get it on video, get it in writing, get your name attached to it. The platforms are going to start pulling first-hand expertise into the answer.

Make sure they can find yours.

A couple of quick things from the Obsero side.

Light mode is live. You can now toggle between dark and light in the platform depending on your mood, your screen, or your meeting. Small change, but effective and easy on the eye.

Prompt discovery is a key growth area. The most common question I get from prospects is some version of "what prompts should I track and what's the search volume?" Prompt volume is nonsense, and treating prompts like keywords misses the point.

Real prompts are full sentences, layered with context, written in the customer's own words. The work isn't keyword research, it's understanding your ICP well enough to know what they're actually asking.

Most of my product time recently has gone into the Prompt Identifier, an MVP that generates the prompts your customers are likely typing into ChatGPT, Claude and Gemini, based on your website, business description and supporting docs.

Grok and Copilot are now live in Obsero. Two new platforms tracked alongside ChatGPT, Claude, Gemini, AI Mode, AI Overviews and Perplexity. Grok has grown from 1.9% to 17.8% US chatbot market share in 12 months. Copilot has 33 million monthly active users and 15 million paid Microsoft 365 seats. Both worth tracking, particularly Copilot for B2B audiences inside the Microsoft stack.

Coming this quarter: an Obsero MCP. This is the one I'm really excited about. You'll be able to plug Obsero directly into Claude (and other clients), run your own analysis inside your own AI workflow, build your own reports, and pull the data into your own BI stack. I genuinely believe this is the future for platforms. The dashboard isn't going to be the destination forever. Companies want their data wherever they're already working. That includes their AI chatbot.

If you're a marketing leader thinking about how to track AI search visibility for your brand, book a demo and let's talk.

That's it from me this week. If something in here landed, hit reply or share with someone who'd find it useful.

Andy