Hey there,

Welcome back to another Don’t Feed The Algorithm, your weekly dispatch from Averi — where we explore the intersections of marketing, AI and the future of work.

This week's biggest number isn't a revenue figure or a user count. It's 85%.

That's how much content ChatGPT retrieves and then throws away before building an answer.

Your page can rank. It can get found. The bot can pull it into its research process. And then — 85% of the time — it decides someone else's content is more useful and leaves yours on the cutting room floor.

That one stat reframes everything happening right now. Google's core update is wrapping up after 19 days of carnage, punishing generic content and rewarding original research. OpenAI just told ad agencies it's expanding ChatGPT ads to all 800 million free users. Anthropic filed explosive court declarations ahead of tomorrow's hearing — with the Pentagon's own emails undercutting the government's case. And NVIDIA told the world AI infrastructure is a $1 trillion market.

But the through-line is selection. Being found isn't enough anymore. You have to be chosen — by the algorithm, by the AI, by the user who clicks through from an answer that cited you instead of your competitor.

This issue is built around that gap. The Fan-Out Audit gives you a 90-minute playbook for mapping the invisible search surface AI uses to build answers. What The Bots Are Reading breaks down exactly how ChatGPT decides which 15% makes the cut. And we're launching v1.0 of Averi Academy — a free certification program to help you become the kind of marketer who doesn't just show up in search results, but gets selected by the systems that are replacing them.

Let's build.

— ZC & The Averi Team

Your content just got more connected.

Here's how it works:

That's it. No manual searching through old posts. No guesswork about what links where.

Why does this matter?

Internal linking is one of the highest-leverage SEO moves you can make — and one of the most neglected.

It builds topic clusters that signal authority to Google, keeps readers on your site longer, and helps AI search engines understand how your content connects.

The problem?

Doing it manually is tedious and easy to forget. So most teams skip it entirely.

Now your content engine handles it.

Every article you publish strengthens the links between everything you've already built.

This is live in your editor now. Open a draft and try it.

— Averi Team

Every section of this newsletter exists because the rules of content marketing are being rewritten in real time. Fan-out queries. AI citation surfaces. GEO. LinkedIn as a citation source. Content scoring. None of this was in anyone's playbook two years ago.

We built Averi Academy to help you learn all of it.

It's a free certification program designed to turn content marketers into AI-powered content marketers — the people who understand how to create content that ranks on Google and gets cited by ChatGPT, Perplexity, and Copilot. You'll learn the frameworks, workflows, and strategies we write about in DFTA every week, structured into lessons you can actually execute on.

Finish the program, earn a certificate you can add straight to your LinkedIn profile.

Whether you're a founder running marketing solo, a content marketer leveling up your skillset, or someone eyeing one of those roles in Don't Let AI Replace You — this is built for you.

👉 Become a Certified AI Content Engineer →

The adoption curve has entered its steepest phase.

What was experimental in 2024 became standard practice in 2025 and is now operational infrastructure in 2026.

The competitive advantage has shifted from "using AI" to "having AI integrated into a systematic workflow." Purpose-built content engines that maintain brand context, generate strategic recommendations, and compound intelligence over time represent the next adoption phase — where the real performance gap opens.

The bots are reading everything. They're just throwing away 85% of it.

A new AirOps study analyzed 548,534 pages retrieved across 15,000 prompts — the largest dataset yet on how ChatGPT actually builds its answers. The headline finding: only 15% of the pages ChatGPT retrieves during answer generation ever make it into the final response as a citation.

That means 85% of content that gets found by ChatGPT gets discarded before the user ever sees it.

But the real story is in how ChatGPT searches. It doesn't just look up your query once. On 89.6% of prompts, ChatGPT fired off two or more follow-up searches internally — what researchers call "fan-out queries." A single user prompt expanded into an average of nearly three total searches. The original 15,000 prompts generated 43,233 queries.

Here's why that matters for your content: 32.9% of all cited pages were discovered only through fan-out queries — not the original search. If you're only optimizing for the keyword someone types, you're invisible on a third of the citation surface. And 95% of those fan-out queries have zero traditional search volume. Your keyword tools literally can't see them.

The authority assumption is dead too. High-DA sites (80-100) had a citation rate of just 15% after retrieval — lower than every other authority tier. Mid-tier sites (DA 20-80) earned 63.6% of all citations and converted retrievals to citations at nearly double the rate of the biggest domains. Topic relevance and content quality are outweighing raw domain authority in ChatGPT's selection process.

Meanwhile, Ahrefs confirmed a parallel shift in Google's AI Overviews: only 38% of AI Overview citations now come from top-10-ranked pages — down from 76% a year ago. The organic-to-AI-citation correlation is structurally decoupling. Ranking #1 used to mean you owned the AI answer too. Not anymore.

What this means for startups: The old playbook was rank → get cited. The new playbook is cover the territory the bots search when they go looking for answers. That means building topical depth across the sub-questions AI systems generate internally — not just targeting the primary keyword. Pages with original data earn 4.1x more citations. Content updated within 30 days gets 3.2x more Perplexity citations. And front-loading your answer in the first 30% of the page captures 44.2% of ChatGPT citations.

The bots aren't rewarding the biggest sites. They're rewarding the most useful ones.

📊 By the numbers:

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.

Anthropic — "A Time and a Place"

Anthropic spent $8 million on a Super Bowl ad to argue that some places shouldn't have ads at all.

The "A Time and a Place" campaign — created with agency Mother and directed by Jeff Low — ran four spots during Super Bowl LX. Each one follows the same structure: a person asks an AI chatbot something deeply personal. The bot starts giving real advice. Then, mid-conversation, it pivots into a sponsored recommendation. A guy asking how to communicate with his mom gets pitched a cougar dating site called Golden Encounters. A man trying to get a six pack gets served an ad for height-boosting insoles. The tagline: "Ads are coming to AI. But not to Claude."

The spots aired the same week OpenAI turned on ads inside ChatGPT's free tier. The timing wasn't an accident.

Why we love it: Mother's CCO Felix Richter nailed the brief in one sentence: "We're using advertising's biggest stage to ask a simple question: does it belong everywhere?" The creative didn't need to explain why ads in AI feel wrong. It just showed it. That's the difference between telling people what to think and making them feel what you already know.

The results were immediate. Claude jumped from #41 to #7 on the App Store. Daily active users surged 11%. Downloads spiked 32%. Sam Altman called the ads "clearly dishonest" on X, which was functionally a second campaign for Anthropic — because nothing validates a competitive position like your competitor losing their composure about it on social media.

And the strategic layer underneath is what makes it brilliant. Anthropic wasn't just running a brand campaign. It was drawing a business model line in the sand: enterprise revenue and paid subscriptions vs. ad-supported AI. The blog post that accompanied the ads declared the company would never show ads in Claude, never let third-party placements influence responses, and never sell user data. That's not a marketing message. That's a positioning wall.

The algorithm could never time a product critique to coincide with a competitor's actual product launch, get the competitor's CEO to amplify the message by attacking it publicly, and turn a philosophical stance into a measurable user acquisition event — all in the same weekend.

The startup takeaway: The best competitive marketing doesn't compare features. It makes the other company's business model the problem.

🔗 Watch the ads

Strategy You Can Steal

The Fan-Out Audit

Block 1: Map Your Fan-Out Surface (25 min)

Pick your 10 highest-priority queries — the prompts your ICP would actually type into ChatGPT or Perplexity. Run each one. But don't just read the answer. Read the citations. Click through every source. Note which pages got cited and which topics they cover. For each prompt, write down the 2-4 sub-topics the AI clearly researched to build its answer. You'll see patterns fast — comparison angles, pricing questions, implementation details, integration specifics that your primary content doesn't address.

Block 2: Build Your Gap List (20 min)

You should now have 20-40 sub-topics the AI explored across your 10 queries. Cross-reference against your existing content library. For each sub-topic, mark one of three statuses: (1) we cover this well, (2) we cover this but it's buried or outdated, (3) we don't cover this at all. Category 3 is your fan-out gap list. Category 2 is your refresh list. Both are costing you citations right now.

Block 3: Prioritize by Citation Leverage (20 min)

Not all gaps are equal. Prioritize by asking three questions for each gap: Does this sub-topic appear across multiple prompts? (High frequency = high leverage.) Are the currently cited sources weak — thin content, outdated, generic? (Weak competition = fast win.) Does this sub-topic connect to a buying decision? (Commercial intent = pipeline value.) Stack rank your gaps. The ones that score high on all three are your first content sprints.

Block 4: Write Your First Fan-Out Piece (25 min)

Pick your #1 gap and draft the skeleton right now. Structure it for citation: lead with the direct answer in the first 30% of the page (that's where 44.2% of ChatGPT citations come from). Use specific numbers, names, and frameworks — pages with 15+ named entities show 4.8x higher citation probability. Include a comparison or evaluation angle if the gap warrants it. Product-discovery and how-to queries have the highest citation rates at 18.3% and 16.9%.

Don't try to write a 3,000-word pillar page in 25 minutes. Write a tight, 800-1,200 word piece that answers the specific sub-question better than anything currently cited. Depth on a narrow topic beats breadth on a wide one — mid-authority sites (DA 20-80) earned 63.6% of all ChatGPT citations. You don't need to be the biggest. You need to be the most useful.

Why this works: Traditional keyword research shows you what humans search for. Fan-out auditing shows you what AI searches for on behalf of humans. Those are two different surfaces — and right now, almost no one is optimizing for the second one. The brands that map their fan-out surface first will own citation share while competitors are still optimizing for keywords that represent less than 70% of the actual search activity happening inside AI systems.

Know of a role that should be here? Reply to this email and we'll feature it next week.

Did You Know? —  95% of the follow-up queries ChatGPT generates while building an answer have zero traditional search volume.

Your keyword tools can't see the surface where a third of all citations are won.

Til next time,