Five Insites

Most of your AI visibility still comes from doing SEO properly, just more often than you'd like

By Insites Marketing Team

This week the AI visibility playbook turns out to look an awful lot like the SEO one, provided somebody actually reads your robots.txt, and the answers themselves are drinking from some fairly unglamorous pipes: Yelp reviews, product feeds and Reddit threads. Elsewhere, "AI-powered" has stopped counting as a position, YouTube is lining review videos up in comparison tables, and texts get clicked far more often than emails, according to a company that happens to sell both. Here are your five.

1. Most of your AI visibility still comes from doing SEO properly, just more often than you'd like

Zyppy spent a few months digging into how AI engines choose what to cite and recommend, and came back with a verdict that will sting anyone who has just had "GEO Strategist" printed on a business card: for most websites, good SEO covers 90% or more of AI visibility, by their reckoning. No separate strategy required, and not even an llms.txt file, at least not yet. The piece splits the job into three goals (being eligible to be read at all, earning citations, and actually getting recommended) and maps every one of them back to levers SEOs have been pulling for years.

The twist is in the housekeeping. Your robots.txt now has to tell GPTBot and Google-Extended, which mostly gather training data, apart from OAI-SearchBot, which does the searching. Plenty of AI crawlers don't render JavaScript, so a page that looks perfectly fine to you can be a blank wall to them. Some sites block Common Crawl without realising it's a seed source for AI, and a broad "block AI bots" rule switched on by a well-meaning tech team can quietly undo a year of marketing effort. The most effective AI strategy this autumn may well be reading your own robots.txt. Zyppy

2. Yelp reviews, product feeds and Reddit threads: the unglamorous pipes AI answers actually drink from

Chris Green has built a four-tier map of the data sources feeding AI search, running from confirmed and live right down to strongly suspected. Some entries are pleasingly concrete. Yelp licenses its reviews, photos and business information to OpenAI, and ChatGPT users can now book a table or join a waitlist without leaving the chat. OpenAI accepts merchant product feeds refreshed as often as every 15 minutes. Google reportedly pays Reddit around $60m a year for access, though Reddit is apparently mulling whether to renew. And since August, Google's AI Mode will book you a hotel and take payment through Google Pay.

Green admits the table may date badly, and the published version still has a few working notes left in, including one suggesting his own weakest row be cut, which is oddly reassuring. He also points out that Yelp isn't a big deal in the UK, so the real question is which local equivalents carry weight where you are. Either way, "is it in Google?" has stopped being the whole question. Your Business Profile, your product feed and your reviews used to be supporting documents. Now they're the source material. Search Engine Journal

3. Put a rival's logo on your homepage. If nothing looks wrong, that's the problem

Arielle Jackson, First Round's Head of Brand and Product Marketing, makes the case that "AI-powered" stopped being a position the moment every credible competitor started saying it too. Her examples show how quickly the ground moves. Cursor launched in October 2023 as an "AI-first" editor, sharpened up around codebase understanding and agents in mid-2024 as Windsurf and Replit Agent turned up, and has since repositioned again around agent orchestration, because the question buyers ask shifted from "which editor?" to "which agent codes with you?". Clay went from "horizontal no-code spreadsheet" to "creative tools for growth teams", then put up billboards featuring a rainbow logo chiselled out of marble, in a category that has otherwise committed wholeheartedly to dark mode.

Her test is pleasingly brutal: swap your logo for a competitor's and see whether your homepage still makes sense. Evian and Liquid Death sell a product that is about as interchangeable as products get, and trade their logos and both ads fall apart instantly. That's the bar. She's also blunt about the tools, pointing out that a language model writes the statistical average of everything ever written, and "if you outsource brand to AI, it's a marketing slop machine." Sounding like everyone else used to take a bit of effort. Now it comes free with the subscription. First Round Review

4. YouTube is lining review videos up in comparison tables, and the creators just became the product page

Announced at Made On YouTube on 23 September, Ask YouTube (the platform's experimental conversational search) is getting shopping features. Search for reviews of a particular product and you may get the videos organised into a comparison table of attributes and categories, shaped around your preferences. Pick the one you fancy and you can carry on asking questions about the product right there on the watch page. The search version is still a limited, US-only, English-only experiment, but the watch-page version is hardly niche: Sundar Pichai says more than 140 million people engaged with it in June.

What YouTube hasn't said is how videos get chosen for those tables, or where the product attribute data comes from. The help page offers relevance, engagement and quality, which is the algorithmic equivalent of "it depends". It also notes that Ask YouTube conversations aren't used to serve ads during the experimental phase, four words doing an awful lot of work. The review that sells a product is increasingly one a creator filmed, summarised by a machine and slotted into a neat little grid beside three rivals. Your own product page has been demoted to background reading. Search Engine Journal

5. Texts get clicked at least four times as often as emails, according to a company that sells both

Dan Oshinsky at Inbox Collective reckons SMS is Robin to email's Batman, and Mailchimp's data backs him up: 12% of people who receive a text click the link, against roughly 2 to 3% for email. Use the two together and click rates come out 97% higher. The best examples are the restrained ones. Punchbowl News sends fewer than three breaking alerts a week, so when its name lights up your phone, you look. The New York Post has beat reporters texting behind-the-scenes Yankees notes to paying subscribers, who text straight back.

Worth knowing the piece is presented by Mailchimp, so the stats deserve the usual squint (and since this very newsletter goes out through Mailchimp, we're squinting at ourselves a little). Oshinsky is at least honest about the catches: that famous 98% SMS "open rate" is really a delivery rate, one emoji shrinks your message from 153 characters to 63, and texting a Dutch number costs 12 credits where Portugal costs two. The rule underneath is older than the channel. Your texts sit next to messages from family and friends, and you stay there exactly as long as you behave like a guest. Inbox Collective

Small business shout-out

This week's shout-out goes to Welby & Wright, an Edinburgh studio where master tiler and artist Craig Peebles hand-makes encaustic tiles, the kind whose pattern is inlaid in coloured clay rather than painted on top, so the design lasts as long as the tile does. He makes limited-edition art tiles (including a Sorrento lemon whose sliced half doubles as a pizza), bespoke nameplates, memorials and even fridge magnets, and he rescues Victorian floors too: if your Minton hallway is missing a tile, Craig will make a reproduction to match the colour and pattern. Know a small business we should feature? Reply and let us know.