Hi Team PR@ctical.
I’ve been talking about this for over a year now, and something seemingly so simple is getting an entire edition dedicated to it.
Prompts are not just questions.
While AI visibility is a valid conversation (heck, we have three service lines built around it), what is happening is so much deeper than that.
In real life, a potential customer is actually pasting competing proposals in Chat GPT or Claude and typing: “narrow these down to the company i should work with.”
Those are tasks. And when the prompt is a task, your content becomes what AI models are using to help people make an actual decision.
Here’s what it means: if the model can’t find anything you’ve created, written or built to respond in these moments, it won’t know what to say (or recommend you at all).
G2 surveyed 1,038 B2B software decision-makers in June.
82% had sourced software recommendations from an AI chatbot in the last two years. About half said the tool’s greatest value came when they were narrowing and comparing, not discovering. And buyers who used an AI chatbot bought from their initial shortlist 80% of the time, against 65% for buyers who didn’t.
Which brings me to my mic drop moment (sort of) : we have to start thinking like a CMO.
A CMO does not plan content for one reader. They plan against segments, stages, and channels, and they do it at scale.
Comms already has that exact skill and has been pointing it at humans literally FOREVER. We call it a pitch: one reporter, one beat, one thing they need. Nobody in PR ever wrote a single pitch, blasted it to 400 journalists, and made it win.
That is the biggest threat to our industry right now with what we consider “AI content.” One authority piece. One imagined reader. Send.
Nope. Won’t cut it.
This is the formula:
Who is asking, what they want, which engine they're in, and whether they're paying for it.
What's new is a fifth: prompt type. Same person, same engine, same day. "What is X" and "narrow these down" are two different jobs, and they pull two different kinds of content.
OpenAI published its own usage data on July 27, and it’s great insight on how people use their platform. Full breakdown in REPORT DROP below, but here’s TL;DR: 25% of marketing-related messages inside ChatGPT come from people who are not in marketing.
Everybody is doing our job in the prompt now. Our job is making sure the raw material they grab is ours.
Two-beat, as always.
Free edition and our paid sister edition.
Paid gets the execution. This week that’s THE PROMPT-TYPE MAP: what to publish for each of the four work prompts, in what structure, and how to tell whether it worked. Plus the Clock, the Receipts, the Report Drop, Platform Moves, the Crossover, and the Delivery Insider.
PS: AI Visibility Certification Day cohort #2 is August 28. This edition and next are the last two before it closes. Details in BUILDING & LAUNCHING.
1. GEO GPT: AI Visibility Engine → asksarah.ai/tools
Paste a URL, get back what the answer engines can actually see and use from that page. I’m re-featuring it for one specific reason this week: run it on a page you wrote to be read and watch how little of it survives as extractable material.
2. The CMO’s Guide to AI Visibility (22 min) → asksarah.ai/learn
Putting this back in rotation on purpose. It’s the closest thing I’ve written to the argument above: how to plan visibility across segments and stages instead of publishing one piece and hoping. If you read one guide before your next planning cycle, read this one.
3. AI Visibility Certification Day, cohort #2 (Aug 28) → stan.store/asksarahevans
One day, live, and you leave with a working audit and a plan you can run on a client Monday. Cohort #1 filled. Two editions left before this one does.
On its Q2 call (revenue $3.6B, up 34%), Shopify said AI-driven traffic and orders to its stores each tripled year over year. President Harley Finkelstein said new-buyer orders from AI channels arrive at nearly twice the rate of other channels, and half of all AI-referred sessions land directly on a product description page, 2.5x the rate of traditional search.
The number that matters most for us is the quiet one: AI searches powered by Shopify’s structured Catalog converted at twice the rate of AI searches relying on scraped page data. Same product, same buyer. The difference was whether the information arrived as clean structured data or as something the model had to scrape out of a webpage.
What to do: take your client’s three highest-stakes claims (price, what’s included, who it’s for) and check whether each one exists as plain, extractable text on the page rather than living inside a designed graphic, a PDF, or a paragraph of positioning. Fix the ones that don’t.
Why it matters: first big public number putting a conversion delta on format rather than message. Being findable and being usable are two different jobs, and nobody is staffing the second one.
One honest caveat: this is Shopify’s earnings call and Catalog is Shopify’s product. Tripling is off an undisclosed base, and Finkelstein was careful to say traditional search still holds roughly a third of storefront sessions. He framed AI as a complement to search, not a replacement. Quote him that way.
🔗 G2 2026 Buyer Behavior Report, Jul 22, 2026
1,038 B2B software decision-makers, fielded June 2026, plus 55 qualitative interviews. Evaluation now eats 40% of the buying journey against 36% for research. And 61% use or plan to use AI agents in buying, though only 9% would let an agent execute a purchase inside guardrails and 2% without pre-approval.
What to do: stop writing only to be found. Write one asset built to survive a comparison: named attributes, real numbers, what you don’t do, and who you’re wrong for. The model needs something to put in the cells.
Why it matters: discovery content gets you into the consideration set. Comparison content keeps you there when somebody types “narrow these down.” Most comms teams have shipped a lot of the first and almost none of the second.
Flag it: G2 is a review marketplace that sells vendor profiles and buyer-intent data. The same report claims review-site citations inside AI answers rose 1.8x, which is a direct advertisement for its own inventory. Treat that stat with suspicion. The shortlist and evaluation-stage numbers are the defensible ones.
🔗 TechRepublic, Aug 6, 2026 · OpenAI Help Center
Atlas stopped working Sunday, August 9. Be precise here: OpenAI announced this on July 9, so the weekend was a deadline arriving, not a surprise. Capabilities moved into the ChatGPT desktop app and the Chrome extension: multiple tabs, downloads, account login, agentic navigation. OpenAI’s stated reason was that it didn’t want people sitting on a discontinued browser that would stop getting security updates. Bookmarks did not transfer automatically. Atlas launched in October 2025 and lasted less than a year.
Same week, ChatGPT Ads shipped real campaign machinery (Aug 7): conversion-optimized bidding in beta for product-feed campaigns, bulk campaign creation, dynamic URL macros, a multi-product carousel in test, and Brazil and Mexico launches. Automatic Advanced Matching turns on for existing pixels August 17.
What to do: open ChatGPT, ask it to research and pick a vendor in your client’s category, and watch what it does with their site. Then flag the Aug 17 Automatic Advanced Matching date to whoever owns privacy review.
Why it matters: the agent moved from a niche browser almost nobody used to the tool hundreds of millions already have open. And paid placement inside the answer surface is now operationally normal. Organic and paid are converging in the same box.
🔗 Search Engine Roundtable, Aug 6, 2026
John Mueller confirmed the general Search performance report includes AI Overviews and AI Mode data. Per Google’s documentation: if a user asks a follow-up inside AI Mode, they are performing a new query, and its impressions, position, and clicks count against that new query. SEOs are already spotting them as odd stubs: “yes,” “yes go on,” “yes, pricing.”
What to do: pull the last 90 days of Search Console, sort by weird short queries, and read the stubs.
Why it matters: this is the cheapest prompt research available to any comms team right now. No new tool, no vendor, no budget line. Those follow-ups are the objections your content isn’t answering.
The AI Accelerator. Increase your brand’s AI visibility in 7 days, guaranteed. → offer.zenmedia.com
Most visibility work starts from what a team thinks people ask. We start from what they actually type, which includes the tasks, not just the questions.
The Accelerator runs the real prompt set for a category, finds where a brand goes missing when somebody is narrowing rather than browsing, and ships the content built to be pulled into that answer. It’s white-label, so agencies can run it under their own brand.
If your team has been asked “are we showing up in AI” and doesn’t have a defensible answer yet, this is the seven-day version of getting one.
📊 This week’s PROMPT-TYPE MAP, RECEIPTS, REPORT DROP, PLATFORM MOVES, and DELIVERY INSIDER are below the fold for paid subscribers →
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.