Marketing Analytics Headlines, Nov. 19

This week is sponsored by SearchMaster, my AI search visibility tool.

November 19, 2026 · Alex Sofronas
Tap an underlined term for a quick definition.

Editorial cartoon: a shopping question in ChatGPT cites a bike shop, while the owner sees mentions and clicks in Business Hub.

A prospect can now talk to your business inside ChatGPT

Last week I wrote about bringing prospects to your agent. Eligible U.S. ecommerce and online service businesses can now publish a customer-facing Business Agent and use it as a destination for ChatGPT ads. Independent of the ads, OpenAI’s Business Hub reports mentions and website clicks. Compare those clicks with your own .

An AI answer can show your business, send someone to your site, or open a separate conversation with your Business Agent. Those are three different actions. Business Hub’s aggregate data is subject to privacy thresholds; it does not give you a transcript of the buyer’s chat. Start by noting your mentions, site clicks and website conversions separately. If you test ads to a Business Agent, keep paid and unpaid traffic apart and decide up front what a qualified next step looks like.

measurement nearly doubled

Muck Rack’s State of PR Measurement found 55% of PR pros had taken steps to measure , up from 28% last year. Among teams measuring, fewer than half use consistent , and only 31% test regularly. Run the same buyer questions every week before calling a change a trend.

The survey retained 571 responses, drawn largely from the vendor’s own contacts. In a separate same-day repeat experiment, Muck Rack found that ChatGPT repeated 41% of cited domains, Gemini 45%, and Perplexity 94% under its test conditions. Those are citation-overlap figures, not brand rankings. Keep the buyer fixed, repeat runs, report each model separately, and save what it cited so you can investigate a change.

Separate brand demand from search discovery

Google’s Search Console API can now split branded and on eligible sites and report clicks from Lens and Circle to Search. Break out the two groups before calling a rise in branded traffic an acquisition win.

Start with separate branded and nonbranded rows, using a comparable lookback. Check whether the lift comes from people already looking for you, or discovery queries that name a problem or category. Keep the visual search slice separate when you can. Those clicks can be useful, but a different mix of search surfaces can distort a simple before-and-after total.

Give automated bids better numbers

Microsoft Advertising is upgrading automated bidding. Its own August data showed 10% more per click at similar for value-based campaigns. Make sure you’re sending real purchase values before changing your bid targets.

Check what counts as a primary conversion and whether its reported value agrees with sales or data. A demo request and a paid order need not carry the same value. Change one material lever at a time so you know whether a shift came from your tracking, your bid target, or the platform. Microsoft’s result is its own data, not a performance estimate you should plug into your forecast.

Your winning bids are a biased sample

Gamrani and Pham’s preprint tackles a blind spot in auctions: you usually only see the when you win. Check how changed as bids moved before calling your paid the market price. Read the PDF.

The paper models a censored auction price and uses randomized exploratory bids to learn more about it. Its results are illustrated with synthetic experiments, not live campaign outcomes. Start with a smaller practical check: group comparable inventory, then plot and paid against bid bands. If you design a follow-up experiment, cap exposure and agree on what you want to learn before changing bids.

Save your Amazon reporting history

Amazon Ads is closing its legacy reporting centers Dec. 31. Export historical reports and rebuild scheduled exports before the switch. Then compare the totals.

Write down which reports feed your dashboards, how far back each one goes, and who receives it. Take a last historical export and recreate recurring reports in the replacement workflow. Run old and new exports for the same dates and window. Compare spend first, then clicks and conversions, and investigate differences before the old workflow is gone.

Track your brand in AI search

I’m testing a weekly email that tracks where your brand gains or loses ground in AI answers and gives you specific changes to try. Want it? Send me your website and one question a buyer actually asks. Check your AI search visibility on SearchMaster today.

Guest spot: Growth Revealed

I joined Don Davis on Growth Revealed to talk about where I’d put a small marketing budget, how bot traffic can make a campaign look better than it is, and why I’d wait for an organic post to work before paying to boost it. Thanks for having me on, Don.

Apple Podcasts · Amazon Music · More appearances

Latest episode

Marketing x Analytics episode with Jason Fishman

Listen on Apple Podcasts · Read the episode transcript

Jason Fishman of Digital Niche Agency told me his first investor campaign raised $2.83 million on less than $100,000, including fees. We got into the part I wanted to unpack: tracking an ad all the way to a closed investment in the .

PDF credit: Yun-Ithry Gamrani and Huyên Pham, Media Trading with Censored Auction Prices. Unchanged Nov. 12, 2026 preprint, CC BY 4.0.

Glossary

Search engine optimization: improving how easily people can find your site in unpaid search results.
Answer engine optimization: making your content easy for search tools to understand and use in direct answers.
Generative engine optimization: improving how often and how accurately generative AI systems surface your brand or content.
How often and how prominently AI assistants mention, recommend, or cite a brand across a defined set of prompts.
Large language model: a model trained on large amounts of text to interpret and generate language.
A reference an AI answer supplies for a source it used or recommends. A brand can be mentioned without being cited.
The question or instruction given to an AI assistant.
A search or AI prompt that names a specific company, product, or brand.
A search or AI prompt about a need, category, or problem that doesn't name a specific brand.
Return on ad spend: ad-attributed revenue or conversion value divided by ad spend. It measures neither profit nor causal lift by itself.
The amount assigned to a completed action, such as an order's revenue or an estimated value for a qualified lead.
Buying and selling ad inventory through automated systems, often using a real-time auction.
The price the winner pays in an auction, determined by the auction's rules and not always equal to the winning bid.
Cost per thousand impressions: total ad spend divided by impressions, multiplied by 1,000.
The share of auctions won out of the auctions entered. Use consistent counts when comparing platforms or reports.
The rules or models used to assign credit for a conversion to marketing touchpoints. Credit is not proof of causation.
Using data to understand marketing performance and decide what to change, test, or fund.
Customer relationship management: the software or process for recording interactions with prospects and customers.
Amazon's tools and reports for examining advertising delivery, spend, and attributed outcomes such as sales.
Visits that analytics classifies as coming through a link on another site. Not all AI-driven visits carry an identifiable referral.
The outcomes that happened because of an intervention and would not have happened otherwise, commonly estimated with a control group.