Cited.

TL;DR this week:

Two years ago the question was whether AI search was real. One year ago the question was whether it would steal your clicks (it did, by 38–58% depending on whose study you trust). This week the question shifted again. Now it's not about traffic loss. It's about whether your brand exists inside the answer at all, and whether you can prove it.

HubSpot AEO Sensor turning on this week is the public infrastructure layer. The 680M-citation study (covered in What The Bots Are Reading) is the diagnostic. And the Anthropic + Klaviyo agentic workflows are the execution layer — your competitors are already wiring Claude into their customer data. The work this week is not "make more content." The work is making content that gets pulled into the answer, and knowing when it does.

I built this issue around that loop. Strategy You Can Steal is a 90-minute sprint to test where you currently get cited and where you don't. Feed The Engine kicks off our content engineering rotation with schema markup. The Algorithm Could Never pick this week is the strangest political endorsement in product placement history.

Let's build.

— ZC

What's the single AI engine you most want to show up in — ChatGPT, Perplexity, Google AI Mode, or Copilot — and why? Best answers go in next week's issue.

This week's term: RAG (Retrieval-Augmented Generation)

RAG is the architectural pattern behind almost every AI search experience you've used in the last 18 months — ChatGPT search, Perplexity, Google AI Overviews, Claude with web access, Copilot, all of them. The model doesn't answer your question from training data alone. It first retrieves a small set of documents from an external index (the open web, Bing's crawl, a private knowledge base), then generates an answer grounded in that retrieved context.

This matters for content marketers because RAG reshapes what "ranking" actually means. In a pure language model world, your content has to be inside the training data to matter, and you have no control over that. In a RAG world, your content has to be inside the retrieval index and structured well enough to win the retrieval step.

That's the entire point of AEO/GEO: you're optimizing for the retrieval layer, not the generation layer. Win retrieval, get cited. Lose retrieval, get ignored. The model itself is roughly the same for everyone.

Next time someone in your Slack says "but ChatGPT just knows this stuff from training," you can say "that's only true for older queries — anything recent or branded runs through a RAG pipeline, and we can optimize the retrieval layer."

The eight biggest stories shaping AI, search, and content marketing this week.

By the end of this sprint you'll know exactly which AI engines cite your brand today, which ones don't, and the single biggest fix for each gap.

Open ChatGPT, Perplexity, and Google AI Mode in three browser tabs.

Run the same five prompts in each:

"Best [your category] tools for startups in 2026,"
"How to do [your top use case],"
"[Competitor name] alternatives,"
"What is [your most important branded term]," and
"Compare [your product] vs [top competitor]."

Screenshot every answer. Note three things per query — whether you were cited, where in the source list you appeared, and what specific URL was pulled. You'll start to see the asymmetry inside ten minutes. Most startups appear in one engine and are invisible in the other two.

Pull up the three most important pages on your site (homepage, top-converting blog post, top product page). For each one, audit four things: Does the page have a single, question-shaped H1? Does the first 60 words contain a direct, extractable answer to that question? Are statistics hyperlinked to primary sources? Is there an FAQ section with FAQPage schema? If you scored less than 4/4 on any page, that's where Perplexity and ChatGPT are quietly skipping you. The Growth Marshal study found that attribute-rich schema earns a 61.7% citation rate versus 41.6% for minimal schema. Schema isn't optional anymore.

Pick the top three pages from Block 2. Add a visible "Last updated May 2026" timestamp to each (in the H1 vicinity, not buried in metadata). Update the lede with a current-quarter framing — "As of Q2 2026, the share of buyers using AI search to evaluate vendors has crossed 60%." Add one new statistic from a source published in the last 30 days. This is the move that wins Perplexity. The 82% freshness rule isn't a suggestion — Perplexity will skip a 90-day-old page in favor of a 20-day-old page even when the older page is structurally better.

Open HubSpot's new AEO Sensor and bookmark it.

Set a recurring 30-minute slot every Monday at 9am to: rerun your five Block 1 prompts, log citation status in a sheet, check AEO Sensor's industry volatility score, and adjust the editorial queue if a competitor newly appears in citations where you don't. Cite-tracking has to become operational, not occasional. We did exactly this at Averi starting in January — we set Monday-morning citation checks across 20 prompts in our category — and our share of ChatGPT citations climbed from 14% to 39% in 16 weeks. None of that came from new content. It came from restructuring existing pages and tightening freshness.

Why this works: The 11% citation overlap finding from this week's What The Bots Are Reading section means you cannot fix this with one piece of content. You have to operate the loop. Audit → diagnose → ship the structural fix → re-audit. The compounding happens at the loop level, not the page level.

Stop running this audit in three browser tabs.

Track citation share, schema completeness, and freshness signals from one workspace → Try Averi

Your website has two files that decide whether AI can read your content. One has existed since 1994. The other was proposed in 2024, and only 5-15% of websites have it. If you don't have both configured correctly, you might be invisible to AI search — or visible in ways you didn't choose.

robots.txt is the gatekeeper. It tells crawlers what they can and can't access. The critical update for 2026: you now need explicit rules for AI-specific crawlers, not just Googlebot and Bingbot. GPTBot (ChatGPT), ClaudeBot (Claude), Google-Extended (Gemini/AI Overviews), and PerplexityBot each have their own user-agent strings. If your robots.txt blocks them — or uses a blanket Disallow: / for unknown bots — AI engines can't read your pages, and pages they can't read don't get cited. The principle is simple: allow crawlers that power AI search products where you can be cited and get referral traffic. Block crawlers that scrape for training without giving you visibility in return.

llms.txt is the guide. Proposed by Jeremy Howard of Answer.AI, it's a plain-text Markdown file in your root directory that tells AI systems what your site is about and which pages matter most. Where robots.txt says "here's what you can access," llms.txt says "here's what we do, here are our most important pages, and here's how to make sense of our content." Think of it as a curated table of contents for machines. Anthropic, Cloudflare, Stripe, and Vercel have already implemented it. Yoast offers one-click generation. Aim for 10-30 URLs — your highest-value, most citable pages — with specific descriptions of what each page covers and what questions it answers.

One thing to do this week: check your robots.txt right now at yourdomain.com/robots.txt. If GPTBot, ClaudeBot, Google-Extended, or PerplexityBot are blocked, unblock them. Then create a basic llms.txt file listing your top 10-15 pages with one-line descriptions and upload it to your root directory. Five minutes of work that opens the door to every AI citation system simultaneously.

Next week: Answer Blocks — how to write 40-60 word extractable paragraphs AI systems can cite cleanly.

Most takeaway pieces on this report will regurgitate 50 stats with a sentence of context each. That's noise.

HubSpot's report is written for marketing teams with $1M+ budgets and 50+ people. The seed-stage founder reading it without translation walks away anxious and confused about which signals to act on.

The 7 takeaways below are the ones that actually shift what a 1–10 person team should do in the next 14 days.

Each section has the stat, the enterprise reading (so you can compare to what your VP friends at bigger companies are doing), and the seed-stage move that compounds.

Let's go.

Two AI engines walk into a query. Only 11% of the time do they cite the same domain.

That's the headline number from a fresh analysis of 680 million AI citations spanning ChatGPT, Perplexity, and Google AI Mode — and it's the most useful single data point any startup marketer can carry into a planning meeting this quarter. Because for the last 24 months, the dominant working assumption has been that "rank in Google → get cited by AI." That assumption is dead. The platforms have different source preferences, different freshness biases, and different volume tolerances, and treating them like one channel is the fastest way to optimize for nobody.

The volume gap alone tells the story. ChatGPT averages 10.42 links per response. Google AI Overviews average 9.26. Perplexity averages just 5.01. Perplexity is the strictest gatekeeper, which is why it's also the most freshness-biased — 82% of Perplexity citations come from content published in the last 30 days, and the presence of a visible year signal in the title or headers ("2026" in your H1, "Last updated May 2026" in the meta) lifts citation rates by roughly 30%.

Layer on top of that what Princeton's GEO study established last year — that structural optimization can boost AI visibility by ~40% — and you start to see what's actually happening. The engines are not running a unified ranking algorithm with a shared notion of authority. They're running different retrieval pipelines with different bias surfaces. ChatGPT skews toward established authority signals (the Growth Marshal study found a 0.334 correlation between brand authority and citation frequency). Perplexity skews toward freshness and structural extractability. Google AI Mode skews toward verifiable, schema-rich content with explicit entity definitions.

Only 17% of AI Overview citations overlap with Page 1 organic rankings. Translation: the link between traditional SEO authority and AI citation is much weaker than anyone selling you "GEO services" would like you to believe. Brand authority helps. Freshness helps. Structure helps. But the three knobs are not the same knob.

What this means for startups: stop optimizing for "AI search" as a category. Pick the engine that drives your buyers (we covered this in last month's GEO Playbook — for B2B SaaS founders, that's almost always ChatGPT first, Perplexity second) and write content shaped specifically for how that engine retrieves. ChatGPT wants depth, citations to authoritative sources, and clear entity definitions. Perplexity wants recency, structure, and extractable answer blocks. Stop trying to win all three with the same article. Specialize.

📊 By the numbers

Check the new HubSpot AEO Sensor for your industry baseline → AEO Sensor, then run the structural audit in the Strategy section below.

We all know the industry loves to pretend it’s got it all figured out.

But here?

We tell the truth.

Every week, we’re asking you to share how things really work behind the scenes. Because getting better starts with getting honest.

Don’t worry, it’s a safe space.

Drop your vote.

Microsoft says only 16% of brands track AI search performance. So let's find out where you actually stand.

If you picked the third option, you're not alone… about a third of the founders I've talked to this quarter are running some version of "prompt → screenshot → spreadsheet" on a Tuesday morning.

It's the most common entry point into AEO. It's also the first thing that breaks when you scale past 20 prompts.

See your AI referral traffic →

Vaseline — "Vaseline Verified"

A 153-year-old petroleum jelly brand just won Social Campaign of the Year at the Ad Age Creativity Awards. Nobody had that on their 2026 predictions list.

Vaseline's team discovered something hiding in plain sight: 3.5 million organic social posts of people using Vaseline in DIY beauty hacks — mixing it with coffee grounds for exfoliation, layering it as a lip mask, using it to set eyebrows. The hacks were everywhere. Most of them were unverified. Some of them didn't work. None of them had the brand's stamp on them.

Instead of creating a campaign from scratch, Vaseline hired actual R&D scientists to test hundreds of these viral hacks in a lab. The ones that passed got a #VaselineVerified seal. The ones that didn't got a clear explanation of why. The brand filmed the scientists testing products, published the results across TikTok and Instagram, and turned their existing user community into the campaign's creative department. The brand didn't generate the content. It validated it.

Why we love it: This is the content engineering playbook applied to a consumer brand. Vaseline found what their audience was already creating, tested it against a credibility standard, and published the results as authoritative content that only Vaseline could produce. The same week that the Ahrefs study shows AI systems selecting for original data and verifiable claims over regurgitated consensus, Vaseline demonstrated what "information gain" looks like in practice: take what already exists, add a layer of original verification that nobody else has, and publish the result. The R&D lab footage isn't just content. It's a citation magnet — specific, verifiable, original, and impossible for a competitor to replicate without literally running the same tests.

Results: 136 million social views in three months. 43% sales uplift. 87% positive consumer sentiment. Titanium Lion and two Grand Prix at Cannes Lions. The campaign extended from 2025 into 2026 because it kept performing — the community kept posting hacks, the scientists kept testing, and the cycle kept feeding itself.

The algorithm could never decide that the most valuable thing a petroleum jelly brand could do with 3.5 million organic posts was hire scientists to prove which ones actually work. That requires reading a community accurately enough to know that validation — not participation, not reposting, not vibing along — is what the audience actually wanted from the brand. The insight isn't "our community is talking about us." The insight is "our community wants us to tell them which hacks are real."

The startup takeaway: Don't create the content. Validate what your audience already created — and publish the proof only your brand can produce.

🔗 See the campaign

FEATURED:

Yes, it's expensive. Yes, it's full of agency holding-company politics. It's also the only week of the year where the entire global creative industry shows up in one room with their best work, and where AI's role in creative production is going to dominate every keynote, every panel, and every after-party. The 2026 program adds the new Cannes Lions Deconstructed stream (a sit-down format unpacking winning work line by line) plus a dedicated B2B summit on June 23. If you're a startup founder building a creative-led brand, Cannes is the place to figure out where AI ends and judgment begins for the next 12 months.

Date: June 23rd, 2026

Location: Cannes, France

Link: Register

Six open roles where AI content, GEO, and AEO show up directly in the title or scope of the work.

The trend across this week's listings is unsubtle… the "Content SEO Manager" title now accounts for 20% of all content marketing listings per Semrush's analysis of 8,000+ postings, and 34% of senior content roles explicitly require AI fluency.

Two years ago "GEO" wasn't on a single job description.

Today it's a salary multiplier — AI-fluent marketers are seeing 20–30% pay premiums over traditional counterparts, with senior specialists clearing $250K total comp.

Getting certified first → Averi Academy

Know of a GEO or AI content role that should be here? Reply with the link and we'll feature it next week.

Did You Know? —  Brands cited inside Google's AI Overviews see 120% more organic clicks per impression and a 41% lift in paid clicks than uncited brands on the same query, per Seer Interactive's 2026 analysis. The AI summary isn't killing your traffic. Being absent from the AI summary is.

Til next time,