Something has quietly changed about how people find businesses, and most owners won’t notice until the enquiries thin out. The classic top-of-funnel search, the “how do I fix a leaking tap” or “best conveyancer near me” query, is now being answered on the search page itself by AI Overviews. The question gets asked, the AI answers it, and nobody clicks through to anyone’s website. That traffic has collapsed, and it is not coming back.
Here’s the twist, and it’s good news if you play it properly. The practitioners tracking this closely report that branded search is up, and that branded leads convert brilliantly. The new buying journey looks like this: someone asks an AI assistant who they should use, gets a name, Googles that name to check the business is real, then rings ready to buy. The AI recommends, the buyer verifies, the phone rings. Your job is no longer to win the generic search. Your job is to be the name the machine hands out.
Where this comes from
We run a research pipeline at Content Hype that pulls apart what marketers and SEOs are publishing, verifies what we can, and bins the hype. The strongest-evidenced thread in it right now is a long conversation between two practitioners who disagree by default. James Dooley runs lead generation sites at serious scale, more than a thousand by his own account. Joy Hawkins runs Sterling Sky, one of the most respected local SEO agencies going, and she has built a reputation on testing claims other people just repeat. When those two independently report the same numbers from their own client data, it’s worth your attention.
One of their observations is worth holding loosely but watching closely. Both reported that the share of leads referred by ChatGPT was falling in their data while leads referred by Claude were rising. That is their reported client data, not a platform announcement, and we treat it as exactly that. The direction matters less than the mechanism it confirms: real buyers are already arriving via AI recommendations, and businesses that aren’t asking “how did you find us” won’t see the shift until a competitor does.
Your captions are ranking assets now
The most practical finding is that Facebook and Instagram posts are being indexed by Google and feeding AI models fast, sometimes within hours in Dooley’s testing. But the rules are specific. It only works on pages and groups, never personal profiles. The rough test is that if you can run an ad from it, it can rank. The first seven to twelve words of a post effectively become its title tag, so you lead with the entity, the actual business name and the actual fact, never with “big news from the team”. The format machines can parse is a plain factual sentence: who, did what, where. “Eastern Roofing installs slate roofing across Melbourne’s inner north” is a sentence a model can file. “So excited for what’s ahead” tells it nothing.
Two quirks matter. On LinkedIn, hashtags kill it, because the hashtag becomes the title, so leave them off there. And the models only read the post text itself, not the comments, and a post doesn’t need a single like to rank. Engagement is a separate game with its own rules. Ranking is decided by what’s literally written in the first sentence.
Sports clubs, as it happens, are unusually well built for this. A club produces named, dated, factual events every single week: results, signings, milestones, awards. An opening line like “Essendon Royals beat Altona Magic 3-1 at Cross Keys Reserve on Saturday” is precisely the entity-first format the machines file away, because it carries the club, the opponent, the score and the venue before it carries any feeling about them. Most clubs bury all four facts under three emojis and a “WHAT A NIGHT”. The clubs that state it plainly become the ones the machines can quote.
The grey stuff, and why we pass
Some of what’s working right now is grey, and we’d rather name it than pretend it doesn’t exist. One tactic doing the rounds is stuffing the word Reddit into page titles and slugs because AI models love citing Reddit. Another is publishing self-promotional pieces on Medium purely to borrow its domain authority. We looked at both properly. They work today. They’re also built on rented land, the platforms and Google have every incentive to torch them, and the shelf life is short. Content Hype doesn’t take clients there. The legitimate version of the Reddit play, actually participating in communities, is a different job done for different reasons.
The boring stuff that lasts
What we did adopt is the white-hat core, and it’s beautifully boring. The things AI models cite are awards, reviews, case studies and third-party mentions, stated in plain factual sentences across lots of surfaces. Dooley’s phrase for the era is that reviews are the new link building, and Hawkins backed the pattern from her own client base. If a machine is going to summarise who’s good at what, your job is to make sure the verifiable evidence exists in machine-readable sentences everywhere it looks. Win an award and the caption isn’t “humbled and honoured”. It’s the name, the award, and the clients the work was for.
One finding from the same conversation deserves a mention because it was independently replicated, which almost never happens in SEO. Dooley claimed that for service-area pages, duplicating your best page across locations outperforms carefully rewriting each one. Hawkins, the resident sceptic, tested it on her own clients and saw the same result. The caveats are real: it leans on existing brand trust and crawl budget, and it does not mean spinning up a thousand pages on a fresh domain, which still fails. But it punctures a decade of “unique content” orthodoxy, and it was proven by someone motivated to disprove it.
Then answer the phone
None of this matters if the resulting enquiry sits in an inbox overnight. By Dooley’s own operations data, around thirty percent of web form leads are never reachable at all: fake numbers, cold feet, moved on. His call centre works to a target of ringing a new lead within twenty-three seconds. Most businesses we know take a day and a half. So here’s the uncomfortable maths: you can spend months becoming the name the AI recommends, then lose the sale in the gap between the form submission and the callback. The cheapest upgrade in this entire piece is answering faster.
Where we’ve landed
Our position is straightforward. The generic search click is gone and chasing it is a waste of your budget. What replaces it is being recommendable: entity-first posts from your page, evidence of being good published in plain sentences, and a front desk that picks up. It’s not glamorous, but it’s durable, and it’s the version of this we’re happy to stand behind in twelve months’ time.
If you want a hand making your content machine-readable, from the first sentence of your captions to the way your reviews and case studies are published, talk to us. That’s the work we do.