Confirmation.

TL;DR this week:

🌐 1B monthly users now run searches inside Google's AI Mode, one year after launch

📊 68% of all AI citations are captured by just 15 domains, with Reddit at #1.

🔄 Google's first AI optimization guide says there is no separate playbook for AI search.

For two years, everyone in this space has been guessing. Is GEO a real discipline or repackaged SEO? Does schema matter for AI Overviews? Should we publish an llms.txt? Are we writing for humans or for retrieval? This week Google walked onstage at I/O, dropped the first official AI optimization guide on May 15, rolled out the May Core Update on May 21, and answered most of those questions in writing. Same content wins both surfaces. llms.txt is not required. Schema helps but is not magic. Top-10 ranking no longer guarantees a citation. Page-3 quality can beat page-1 thin content.

The 5W AI Platform Citation Source Index, published the same week, confirmed the other half of the story. The top 15 domains capture 68% of every AI citation across ChatGPT, Perplexity, Gemini, Claude, and AI Overviews. That is more concentration than PageRank ever produced. The pie is shrinking. The players who get in early get in for years.

Inside this issue: the citation pool data, a sprint to audit your own entity density, and a Feed the Engine that actually mechanically explains why Answer Blocks still earn you AI lift.

Let's build.

— ZC

Where did you waste budget this year on GEO tactics that turned out to be SEO basics in a hoodie? Best answers go in next week's issue.

This week's term: Citation Threshold

Citation threshold is the internal scoring bar an LLM uses to decide whether a retrieved passage is good enough to include in its generated answer with a visible source link. Every modern AI search engine retrieves many more passages than it actually cites. ChatGPT may pull 50 candidate passages from its retrieval system for a single query, but cite only 4 of them in the final answer. The cutoff is the threshold. Below it, the passage influences the generated text but is never shown to the user as a source.

This matters for content marketers for one reason. The 5W index this week showed that ChatGPT mentions brands 3.2 times more often than it cites them. That gap is the threshold gap. Your content is being read and used as input. It is just not being attributed because something about it does not clear the bar — usually entity ambiguity, missing structured data, or a passage that does not stand alone as a clean factual unit. Improving citation rate is mostly about pushing more of your passages above the threshold, not about producing more passages overall.

Next time someone in your Slack says "ChatGPT keeps mentioning us but never linking us," you can say "you're hitting the citation threshold gap — your passages are influencing answers but not clearing the attribution bar, which means your entity signals or structured data are too weak for the model to confidently source you."

Shopify and HubSpot CMS publishing are live inside Averi.

Two integrations. One thesis: the content engine shouldn't stop at the draft.

When we launched CMS publishing with Webflow, Framer, and WordPress, the feedback was immediate — founders stopped context-switching between their content workflow and their publishing workflow. Strategy, drafting, scoring, and publishing lived in one place. The bottleneck between "approved" and "live" disappeared.

But we kept hearing the same question from two audiences we couldn't serve yet: "We're on Shopify" and "We're on HubSpot."

Now you're covered.

Shopify CMS. Publish blog content directly from Averi to your Shopify store. Draft in Averi, score it, approve it, and push it live — without logging into Shopify admin, without copy-pasting into the blog editor, without losing formatting. For e-commerce founders running content as a growth channel alongside product pages, this closes the gap between content strategy and storefront.

HubSpot CMS. Publish directly to HubSpot's blog and landing page CMS from inside Averi. This one matters especially this week — HubSpot just launched its own AEO product and shifted to outcome-based agent pricing. If you're on HubSpot's marketing stack but want your content strategy, scoring, and publishing workflow powered by Averi's content engine, you no longer have to choose. Use both. Averi handles the strategy and the draft. HubSpot handles the CMS and the CRM. One publish button connects them.

The content engine doesn't care where your site lives. It cares that your content gets from strategy to live without a single manual handoff slowing it down.

Connect your CMS →

What it is: Your pages have the answers. They're just buried in flowing prose that AI systems can't extract. This sprint takes your five highest-value pages and restructures the key passages into citation-ready answer blocks — without rewriting the full article. Five pages, 90 minutes, zero new content required.

Time commitment: 90 minutes, 4 blocks

Block 1: Pull Your Citation Candidates (15 min)

Open Google Search Console. Filter to the last 90 days. Sort by impressions, descending. Export your top 20 pages. Now cross-reference: for each page, search the same query in ChatGPT and Perplexity. Are you cited? If yes, screenshot it — that's your baseline. If no, that page has high impressions but zero AI citation. Those are your candidates. Pick the five with the highest impressions and no AI citations. These pages are getting seen by Google but ignored by the machines — the gap is structural, not topical.

Block 2: Find the Buried Answers (20 min)

For each of the five pages, identify the single most important question the page answers. Not the keyword — the actual question a human would type into ChatGPT. Then find where in the article you answer it. In most cases, the answer is there but it's woven into a 200-word paragraph, nested under the third subheading, and preceded by two paragraphs of context. That's why AI skips it. Mark the passage. Copy the core claim into a separate doc. You now have five answers that exist on your site but are invisible to citation systems.

Block 3: Rewrite as Answer Blocks (40 min)

For each of the five answers, rewrite it as a standalone 40-60 word block using this formula: direct factual statement → specific stat or named source → self-contained closing sentence. The block must make complete sense if someone reads it with zero context — because that's exactly how AI reads it. Place each answer block directly below the H2 that matches the question. If your H2 isn't phrased as a question, rewrite it as one — question-based H2s with direct answers in the first sentence below are the highest-leverage AI Overview citation pattern. When we restructured our top five pages at Averi using this exact format, three of the five picked up new ChatGPT citations within 30 days without any change in Google rank.

Block 4: Add the Citation Multipliers (15 min)

For each page, add three things that increase the odds your answer block gets selected over a competitor's. First, a named source attribution inside the block ("according to [source], [year]") — AI systems cite passages with explicit attribution at 37% higher rates. Second, FAQ schema at the bottom of the page with 3-4 additional questions, each answered in the same 40-60 word block format. Third, update the "Last Updated" date — 76.4% of ChatGPT's top-cited pages were updated within 30 days. Push the changes live and submit the URLs through IndexNow. Set a calendar reminder to re-run the ChatGPT and Perplexity searches in 30 days and compare.

Why this works: The Omniscient Digital study of 23,000+ citations showed that AI systems don't cite pages — they cite passages. The passage has to lead with the answer, contain a verifiable fact, and work as a standalone unit. Most content has the right answers in the wrong structure. This sprint doesn't change what your pages say. It changes how they say it — and that structural difference is the gap between a page AI reads and a page AI cites.

Run the sprint inside Averi → Content Scoring flags your extractability gaps

When ChatGPT cites your page, it doesn't cite the page. It cites a passage. A 40-60 word chunk that answers a specific question, makes sense when pulled out of context, and contains at least one verifiable fact. That chunk is an answer block — and if your content doesn't have them, AI systems will skip your page for one that does.

The Omniscient Digital study of 23,000+ AI citations found that the passages getting cited share a consistent structure: they open with a direct factual statement, include a specific number or named entity, and close in a way that doesn't require the reader to have seen the paragraph above. They're self-contained. That's the mechanical requirement. AI retrieval systems extract chunks, evaluate them independently, and decide whether each chunk is worth citing on its own merits. A brilliant paragraph that depends on three prior paragraphs for context won't get selected — because the system never reads it that way.

The format is simple. Lead with the answer. Add one stat or named source. Close the thought.

Here's what a good answer block looks like:

"B2B startups that publish 12 or more optimized pages within a 90-day window see up to 200x faster AI visibility gains than those publishing four, according to Brandi AI's 2026 study. The threshold works because AI citation systems evaluate topical depth across a cluster, not keyword coverage across a domain."

That's 47 words. It answers a question ("how much content do I need to get cited?"), includes a specific stat with a source, and makes complete sense if you read it with zero context. An AI system can extract it, cite it, and link back to you.

One thing to do this week: open your three highest-traffic blog posts. Find the section where you answer the reader's core question. Rewrite that section as a standalone 40-60 word block that leads with the answer, includes one stat, and doesn't depend on anything above it. Put it directly below the relevant H2. That's one edit per page, ten minutes each, and it's the difference between a page AI reads and a page AI cites.

Next week: Author Entity Signals — why bylines, bios, and LinkedIn profiles are now ranking factors.

Score your content's extractability → Content Scoring in Averi

The content engineer role didn't just emerge. It split. There are now two ways to put one inside your company.

The first is the version Jasper, AirOps, and every recruiting agency is selling: a $161,000 base salary hire, fully loaded to roughly $200,000 in Year 1, who spends the first six months building the infrastructure your content production will run on.

The second is a $99/month subscription that ships the same six-month build state on day one, then runs the same six functions continuously for less than 1% of the loaded annual cost.

We've called this The AI Content Engineer.

It is not a tool that helps a content engineer do their job. It is the role itself, instantiated as software.

AI is reading the same fifteen websites, over and over, on behalf of a billion users a month.

That is the consolidated picture from the 5W AI Platform Citation Source Index 2026, released May 1 and updated again this week with new ChatGPT data showing Wikipedia and Reddit together drive over 25% of all ChatGPT citations in the United States. The index synthesized 680 million individual citations across ChatGPT, Google AI Overviews, Perplexity, Gemini, and Claude, drawn from six of the largest published citation studies between August 2024 and April 2026. It is the most authoritative cross-engine view of who gets cited that has been published to date.

The top-line finding: the top 15 domains capture 68% of all consolidated AI citation share across the five major engines. That is a level of concentration far above what Google PageRank ever produced inside the open web. For comparison, the top 15 domains in classic Google organic results captured roughly 25% to 30% of branded query traffic in 2019, the last clean baseline year before AI Overviews. Citations have collapsed into a much narrower funnel.

Reddit ranked #1 across every engine, cited at roughly 40% frequency. Wikipedia is the clear leader inside ChatGPT specifically, accounting for between 26% and 48% of its top-10 citation share depending on category. The Wall Street Journal, New York Times, and Bloomberg do not appear in the top 20 for any major engine. Traditional editorial authority and AI citation authority are now two different things.

The second half of the story is volatility. A separate analysis from DesignRush published this week found that ChatGPT 5.4 cited a particular brand 56% of the time on a controlled query set, while ChatGPT 5.5 cited that same brand only 8% of the time on the same queries. A single model version bump produced a 52-point swing in visibility. The Swiss-German measurement study from arXiv tracked daily AI citations across four engines over a 45-day window and found similar instability between prompts and across days. Brand visibility inside any single platform-version pair is unreliable.

What this means for startups: stop optimizing for citation share inside one engine. The pool is small enough now that you need to play wherever the pool is. That means a Reddit presence that isn't spammy, a Wikipedia-eligible entity footprint, structured data that survives version churn, and content that is citation-worthy at the passage level rather than ranking-worthy at the page level. The brands winning AI visibility in 2026 are the brands whose entity is cleanly defined across multiple high-citation domains, not the brands writing the most blog content.

This is why we keep saying GEO is a distribution problem dressed as an SEO problem.

📊 By the numbers:

Chili's — "Fast Food Financing" → "Big Crispy Food Court"

In April 2025, Chili's opened a fake payday loan shop next to a McDonald's in Union Square. The garish storefront screamed "Fast Food Prices Are Out of Control!" and "Finance Your Fast Food Today!" Visitors who got "approved" received a $20 gift card and were led through a vault into a speakeasy where Chili's served its Big QP burger — a direct shot at the Quarter Pounder. Three-hour lines. Trending on X within hours. Over 6 billion media impressions. No paid amplification.

Then they came back. Same intersection. Same McDonald's. April 16, 2026.

This time it was the "Big Crispy Food Court" — a mock courtroom where the public served as jury in the case of Chili's Big Crispy chicken sandwich versus fast food's equally priced but smaller offerings. Themed artwork of chickens and sandwich-loving jurists. A courtroom enactment. Branded merch. The pop-up ran one day and generated coverage in Ad Age, Campaign, Marketing Dive, and dozens of food and business outlets.

A week later, Ad Age named Chili's Brand of the Year at its 2026 Creativity Awards.

Why we love it: Chili's entire marketing strategy is built on one answer: a sit-down meal with real plates and actual service now costs the same as a drive-through combo meal. That's it. One claim. Everything else — the pop-ups, the product names, the $137 million marketing budget (up from $32 million in 2022), the CMO George Felix being willing to plant a fake store next to his biggest competitor — is just finding increasingly creative structures to deliver that single answer in a way people can't ignore.

The parallel to this week's Feed the Engine is almost too clean. An answer block is a single claim, stated clearly, structured to be extracted and cited on its own. Chili's built a two-year campaign strategy that operates exactly like an answer block: one fact, impossible to miss, self-contained, and repeatable in a single sentence. "Chili's is cheaper than McDonald's" is the marketing equivalent of a 47-word extractable passage. The insight fits in one line. The execution makes it undeniable.

The algorithm could never decide to open a fake payday loan shop next to your biggest competitor, come back to the same corner a year later with a courtroom, and trust that the same one-line argument would land twice — harder the second time — because the audience's frustration had only deepened. That requires reading the economic mood of your customer in real time, not optimizing a brief from last quarter's data.

The startup takeaway: If your value proposition can't be stated in one sentence and proven with one stunt, it's not sharp enough yet.

🔗 See the Food Court coverage

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.

Three of this week's six listings are at companies that have never publicly hired for AEO/GEO before — Stripe, Experian Health, and ADT. The discipline is moving from "agency specialty" to "in-house owned function" at large enterprise brands. Average AEO compensation is now $86,556 in the US per ZipRecruiter, with senior roles clearing $174K. This is a market that priced itself in over the past 12 months.

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? —  The average AI Mode search on Google is three times longer than a traditional Google search, and follow-up queries inside AI Mode grew over 40% month over month in the US. People aren't searching less under AI — they're searching deeper, and they expect to be answered, not redirected.

Til next time,