Hey friend, how’s it going? 🌻
I rated seven new and old content formats out of 10. Purely through the lens of AI search and how it’s shifting.
Let’s get into it 👇
📌 TL;DR
Informational TOFU scored a 3. Ahrefs found 99.9% of AI Overview triggers are informational intent, so there’s nowhere to hide.
MOFU listicles scored 6.5, and I had the reason the wrong way around. It isn’t domain authority holding you back. 35% of the “best” lists ChatGPT cites sit on low-authority domains. It’s freshness.
BOFU scored 7.5, and the score has an expiry date. Its protection is a Google product decision, not a content moat.
LinkedIn is the surprise of this list. Founder posts and newsletters both scored 8 because ChatGPT and AI Mode cite individual creators 59% of the time rather than company pages.
Thought leadership scored 6, which is an average, hiding two different answers. With proprietary data, it’s an 8. Without it, a 4.
Original research scored 9, the only 9 here. Primary research pages are 3.3x more citation-dense than everything else.
1. Informational TOFU articles: 3/10
You know the article. “What is sales outreach?” “How does X work?” The kind you built your whole content engine on because it brought in volume, and it fed the funnel, and everyone agreed that was the job.
Ahrefs analyzed 146 million SERPs and found that 99.9% of keywords that trigger an AI Overview have informational intent. And when that AI Overview appears, the page sitting at position 1 loses 34.5% of its click-through rate.
So the traffic argument for TOFU is gone. That’s the 3. But here’s what stops it from being a 0, and this is the part I’d have missed if I hadn’t gone digging.
Kevin Indig looked at nearly 4,000 domains across four AI engines and found something uncomfortable: 74.9% of domains get cited, but only 38.3% get mentioned by name. He calls them ghost citations. Your page is in the answer. Your brand isn’t.
Which means your TOFU library might be feeding AI answers right now while your analytics show you nothing at all.
If you kill your TOFU today because the clicks died, you’ll disappear from AI answers in six months and never connect the two. It’s a 3 for traffic. It’s maybe a 6 for being in the room.
2. MOFU articles and listicles: 5.5/10
I had this at 5.5, and my reasoning was completely backward. Let me show you the thing I got wrong, because you might be carrying the same assumption.
My assumption: listicles are middling because G2 and Capterra have domain authority I’ll never touch, so they take the citations, and I don’t.
Glen Allsopp ran a study for Ahrefs on 750 top-of-funnel prompts across software, products, and agencies, manually categorizing 26,283 source URLs. “Best X” blog lists were the most prominent page type at 43.8%.
Here’s the part that killed my assumption. In that study, G2 and Clutch-style pages are counted separately as “non-blog lists.” They’re not the ones taking the 43.8%. Ordinary blog lists are.
And then this: 35% of the cited best lists were on low-authority domains. Some of them, in Glen’s own words, questionable.
So the gate isn’t a domain authority. I was wrong about that, and if you’ve been telling yourself you can’t compete with G2, you might be wrong about it too.
It is freshness. Of 1,100 cited lists he checked, 79.1% had been updated in 2025, and 26% had been touched in the previous two months alone.
So why 6.5 and not higher?
Because being on other people’s lists still beats writing your own, and the same study found a clear trend towards brands positioned near the top of a list getting recommended. That’s an outreach problem, not a writing problem. Your content team can’t fix it.
3. BOFU articles: 7.5/10
I was ready to give this an 8, and I still nearly did.
The case for BOFU is real.
Ahrefs found that shopping is the lowest AI Overview category at 3.2%, and navigational queries trigger just 0.09% of the time.
Seer puts transactional at 5%.
And the intent quality holds up: Search Engine Land’s 13-month analysis across their client base found that LLM referrals convert at roughly 18%, higher than paid search, SEO, and PPC.
So BOFU is protected, and the people who read it are the people who buy. Good.
BOFU Content E-book: Get my updated BOFU Content e-book that covers three new chapters on BOFU content beyond alternatives and comparisons, BOFU attribution, and AI workflows for BOFU content.
But I had the reason wrong, and I think a lot of us do.
I’d been telling myself BOFU is safe because it’s hard to summarise. Product-focused, specific, too much nuance for an AI Overview to flatten. That’s a nice story, and it isn’t true. An AI Overview could summarise your BOFU page tomorrow. It doesn’t, because Google has decided which intents get an AI Overview.
That’s not a content moat. That’s a policy moat. And policy moves.
It’s already moving. Semrush tracked 600,000 keywords between November 2025 and April 2026 and found the share of commercial SERPs carrying an AI Overview grew 71%. Their longer study of 10 million keywords shows that transactional AI Overview triggers climb from 1.98% to 13.94% over 12 months.
7.5 today. Ask me in March.
4. Founder LinkedIn posts: 8/10
I had this one at 7.5 in my notes. The data pushed it up.
Profound analyzed 1.4 million citations between November 2025 and February 2026. LinkedIn moved from around #11 to around #5 in ChatGPT’s domain rank, more than doubling its citation frequency. It was the biggest domain authority shift they tracked all year. For professional queries specifically, LinkedIn is now the most-cited domain across ChatGPT, Gemini, AI Overviews, AI Mode, Copilot, and Perplexity.
But the number that matters for founders is this one: ChatGPT Search and Google AI Mode cite individual creators 59% of the time over company pages.
Your founder’s post outranks your company page. In the answer engine. On your own category question.
There’s even a length spec buried in there. Feed posts perform best between 50 and 299 words, which is roughly the post you’d write anyway if you weren’t trying to game anything.
Why not a 9?
Because it’s rented land, and I don’t want to pretend otherwise. Profound’s analysis leans partly on a synthetic prompt basket, so read it as direction rather than gospel. And a platform that reprices this fast in your favor can reprice just as fast the other way.
5. LinkedIn newsletters: 8/10
I wanted to give this a 10. I’ve been telling people it’s a 10.
Here’s why it isn’t.
The supporting data is genuinely strong. Profound found that posts, long-form articles, and newsletters together rose from roughly 27% to about 35% of LinkedIn’s citations inside ChatGPT, while profile pages collapsed from nearly 34% to under 15%. Answer engines are moving away from static identity and toward published expertise. Articles in the 500 to 2,000 word range draw the most citations.
Your podcast episode, your webinar, your customer interview, your case study, nobody reads. All of it has a second life as something an AI engine can cite weeks later.
But newsletters specifically are still a rising tier-3 source. They’re not top-five on any platform yet. The citation lift is coming from LinkedIn, the domain, not from newsletters, the format. Giving it a 10 would mean claiming credit for the platform’s win.
And one thing I’ve said before that I need to walk back: I’ve told people that company pages don’t get visibility. In AI search, that’s inverted on at least one engine. Perplexity cites company pages 59% of the time. Founder-first is right for ChatGPT and AI Mode. It isn’t a universal law.
6. Thought-leadership articles: 6/10
This score is an average, and the average is hiding the whole point.
I’d been saying that thought leadership is protected because it’s hard to AI-overview. Except that most thought leadership is informational in intent, and informational intent is the single most AI-overlooked category on the internet. So that protection was never real.
What actually protects it is whether there’s something inside that a model can’t reconstruct from ten other pages. Ahrefs, studying 1.4 million ChatGPT prompts, found that pages whose titles and URLs matched ChatGPT’s narrower sub-questions got cited more than pages that only matched the broad prompt.
Read that again if you write thought leadership. You don’t lose on quality. You lose on framing. A piece that never declares which question it answers is invisible to a model that’s busy breaking a prompt into six smaller ones.
So the real scores are: Thought leadership carrying proprietary data or your own specific numbers: about an 8. Thought leadership that’s a well-argued opinion with nothing in it only you could know: about a 4.
Averages out to 6. But you’re not writing the average. You’re writing one or the other.
7. Original research reports: 9/10
The only 9 on this list, and the data isn’t close.
Growth Memo found that primary research pages averaged 11.3 citations each. Everything else averaged 3.4. That’s 3.3x the citation density. And the scarcity is the whole story: only 8 pages out of 301 qualified as genuine primary research, and those 8 pages took 8.4% of all citations in the set.
Rare, then over-indexed. That’s the asset.
Original data gets you eligible. Structure gets you cited. Those are two different jobs, and most research reports only do the first one.
A good read on this: How often does AI actually change a buying decision?
So what do you actually do with this
If I’m reading my own scores back honestly, the pattern isn’t “AI search killed content.” It’s narrower than that.
AI search punishes content that could have been written by anyone. It rewards content that could only have been written by you, and only if you formatted it so a machine can lift it.
That’s it. That’s the whole shift.
📚 Check out my digital products & services purchased and loved by 250+ writers and marketers
AMA: SaaS Content Marketing | 45 Minutes: Strategy, writing, analytics, conversion, SEO, AEO, nothing is off limits. Share your website or portfolio in advance, and I’ll come prepared with specific answers rather than generic advice.
SaaS Content Marketer Handbook: 1-YEAR handbook for content marketers joining a product startup, includes realistic goals, execution steps, and all the templates and frameworks you’ll need.
AI+Human Content Playbook: For Content Marketers only. Essentially, for those under pressure to produce a large volume of content, it limits their ability to learn and experiment with other aspects of content marketing.
💡New update: A Claude skill.md file to accelerate your writing workflow! No price change.
Say Hi on LinkedIn 👋
Onwards and upwards,
Sreyashi