DEV Community: Yousif Alias

35% of Consumers Now Send AI Agents to Customer Service

Yousif Alias — Wed, 28 Oct 2026 14:29:40 +0000

What did the Capacity survey find?

On October 27, 2026, Capacity published The Agentic Consumer Report, a survey of 1,000 US consumers. It found that 35% have already sent their own AI agent to handle a customer service request, and 44% say an AI tool has contradicted a company representative or the company's own policy.

The other numbers in the release on PR Newswire:

55% have used some AI tool, an agent or a chatbot like ChatGPT, to help with a customer service request.
Among Gen Z and Millennials, 49% have used a personal AI agent for customer service.
34% say AI makes them ask more specific questions when they deal with a company.
23% would ask a representative to explain or cite company policy because of something AI told them.
When AI and a company representative disagree, consumers are about equally likely to believe either one.

Capacity sells customer service automation, so treat this as vendor research. The release doesn't give field dates or name the panel provider, a gap The Agile Brand Guide also points out. I still think the direction is right, because I hear it on our own calls.

Why does it matter that 44% say AI contradicted a company?

A big share of your customers now show up with a second opinion already in hand, and that second opinion is often wrong about you. If ChatGPT told a lead your service costs $99 or that you offer free cancellations, your rep is now arguing with a source the customer trusts as much as the rep.

The AI usually isn't lying on purpose. It fills gaps. If your website doesn't say what you charge, how refunds work or which areas you serve, the model guesses from reviews, competitors and old pages. Then the customer calls you quoting the guess.

Capacity's founder and CEO, David Karandish, said in the release that businesses "need to rethink how they measure success, moving beyond speed and efficiency." I agree with half of that. Speed still matters a lot. A fast answer now also has to match what every other source says about you.

What does this mean if you run ads, answer leads or close sales?

Your lead may have researched you with AI before the first call, and some will have an AI agent message or call you on their behalf. Whoever answers, a person or your own AI, has to give the same answer every time, quickly, and from information you control.

Here's where it shows up:

Where	What changes	What to do
Ads	People click, then ask ChatGPT about you before they fill out the form.	Put pricing ranges, service areas and guarantees on the landing page in plain words.
Answering leads	Some first messages will come from a customer's AI agent, with very specific questions.	Answer within minutes, with the same facts your site shows.
Sales calls	Prospects quote things AI told them, some of it wrong.	Ask early what they've already read, then correct it calmly with your written policy.
Support	23% may ask your rep to cite the policy.	Keep refund, cancellation and warranty terms on one public page and link to it.

Speed still wins the deal. The Harvard Business Review study of lead response found that companies that tried to reach a lead within an hour were nearly seven times more likely to qualify it than companies that waited longer. An AI agent shopping for a plumber won't wait until Monday for you to text back. I wrote more about that in why leads go cold.

We built the KenjiAI agents to answer from your own business info for this reason. Conversation AI handles texts and chat, and the Voice AI agents pick up calls and book appointments. They reply in seconds, day or night, with the same prices and policies your team uses, so a customer's AI gets the right answer the first time. We've run ads since 2013 for 500+ businesses, and the leads that convert are the ones that get a fast, consistent reply.

What should you fix this week?

Write down the five facts customers ask about most, publish them on your website in plain language, and make sure your team, your texts and any AI agent you use say exactly the same thing. That page becomes what AI tools read and what your reps point to.

Ask ChatGPT, Gemini and Perplexity what you charge, what you offer and what your refund policy is. Write down every wrong answer.
Fix each wrong answer at the source: your website, your Google Business Profile and your service pages.
Put one policy page live (pricing ranges, cancellations, refunds, warranty, service area) and link to it in your text replies.
Check how fast you answer a new lead today. Text your own business from another phone after hours and time the reply.
Have your closers ask "what have you read about us so far?" in the first two minutes, so wrong info comes out early.

Start with step 1 tonight. It takes ten minutes, and you'll see what your next lead already believes about you.

FAQ

What percentage of consumers use AI agents for customer service?

Capacity's October 2026 survey of 1,000 US consumers found 35% have used their own AI agent to handle a customer service request at least once. Among Gen Z and Millennials the number is 49%. Overall, 55% have used some AI tool to help with a service request.

Why would AI contradict a company's policy?

AI tools fill in missing information by guessing from reviews, competitor pages and old content. If a business doesn't publish its prices, refund terms or service area clearly, the AI often gets them wrong. In Capacity's survey, 44% of consumers said AI had contradicted a company representative or its policies.

How can a small business make sure AI tools describe it correctly?

Publish your key facts in plain language on your website and Google Business Profile: pricing ranges, services, service area, refund and cancellation terms. Then ask ChatGPT, Gemini and Perplexity about your business and fix any wrong answers at the source. Keep your team and any AI agent you use working from that same information.

Should a business answer a customer's AI agent the same way it answers a person?

Yes. Give the same facts at the same speed. An AI agent acting for a customer often asks very specific questions and moves on quickly if nobody replies, so a fast, consistent answer from your own written policies is what keeps the job.

Originally published on kenjiai.com. Published by the Media Traffics | KenjiAI team.

OpenAI Data: Writing Fell to 21% of ChatGPT Requests

Yousif Alias — Tue, 27 Oct 2026 14:38:33 +0000

What did OpenAI release on October 26?

OpenAI released version 3.0 of its Signals data on October 26, 2026, along with a report called ChatGPT’s expanding world of work. The numbers show people now use ChatGPT more for advice than for writing. Writing fell from 34.1% of consumer requests in July 2024 to 21.4% in September 2026.

The data comes from automated classification of 1 million consumer ChatGPT messages a month, sent by adults, over 27 months. No person reads the messages; a model sorts them into topics. PPC Land went through all 24 spreadsheets and published the topic breakdown on October 27. OpenAI's own page focused on work: it maps work requests to 165 job activities across nearly 100 countries, and Blockchain.news reported that the ten most common activities fell from 56.0% of work messages in Q1 2025 to 51.1% in Q3 2026. Use is spreading out.

One limit up front. The files cover individual ChatGPT plans only. Enterprise accounts and Codex are excluded, so business use is understated.

What are people asking ChatGPT now?

Practical guidance is the largest category at 28.1% of requests, and seeking information grew to 20.2%. Practical guidance means how-to advice, tutoring, ideas and health questions. Seeking information covers facts, recipes and questions about products or services you can buy.

Topic	July 2024	September 2026	Change
Practical guidance	26.1%	28.1%	+2.0 pts
Writing	34.1%	21.4%	-12.7 pts
Seeking information	16.7%	20.2%	+3.5 pts
Multimedia (images and other media)	2.0%	8.8%	+6.8 pts
Technical help (coding, math, data)	13.5%	4.6%	-8.9 pts

Source: OpenAI Signals v3.0, as tabulated by PPC Land. The direction matches the NBER paper OpenAI published with Harvard and Duke economists in 2025, which found non-work messages had grown past 70% of consumer use.

The age split matters if you sell to homeowners and business owners. In September 2026, seeking information was 26.7% of requests from users aged 55 to 64 and 24.2% for ages 45 to 54, against 17.3% for 18 to 24 year olds. Older buyers use it most like a search engine.

Why should a business owner who runs ads care?

Your customers are asking ChatGPT the questions they used to type into Google or ask a salesperson: how much a roof repair costs, which med spa treatment fits, what to ask a contractor. Whoever wrote the clearest answer gets quoted. A business with thin pages gets skipped.

That is also where OpenAI is taking its ad business. On October 21, OpenAI said ChatGPT ads now run in 89 countries and more than 100,000 advertisers have run campaigns. Product and service questions sit inside the seeking information category, which is the closest thing in this data to search intent. OpenAI does not split out how many requests are product questions, so nobody outside the company knows that number yet.

I've been running ads since 2013, and every channel shift looks the same at the start. The early people get cheap attention because the answers and the ads are thin. That window is open right now for local service businesses.

What should you do this month?

Write down the ten questions customers ask before they buy, and publish a short, direct answer to each one on your site. Then make sure the lead that comes out of that research gets answered fast, because a prospect who just read three AI answers about you will call or text the first business that looks credible.

Pull the top ten questions from your call logs, texts and sales calls. Use the customer's words, not yours.
Answer each one in the first two sentences of its own page or section, with a real price range, timeline or rule where you can give one.
Add the proof AI tools look for: reviews, hours, service area and real photos. Our guide on getting more Google reviews on autopilot covers the review side.
Ask ChatGPT those same ten questions about your city and see who gets named. That is your competition now.
Check how fast you answer the leads you already get. If calls go to voicemail after 5pm, the research your prospect did is wasted. Our missed call calculator puts a dollar number on it.

Where KenjiAI fits

KenjiAI answers the call or text that comes after the research. Our Voice AI agents pick up every inbound call, qualify the caller and book the appointment on your calendar, and Conversation AI does the same over SMS and chat. Review requests go out automatically after each job, so the next person who asks ChatGPT about you finds proof. We've built this for 500+ businesses.

What is still unclear in the data?

Three things. The data excludes business accounts, so work use is undercounted. The categories come from a model reading messages, not from users describing themselves. And OpenAI does not publish how many requests are about buying something, which is the number advertisers want most.

Watch for two things next: whether OpenAI splits product questions out in a future Signals release, and how ChatGPT ad prices move as more advertisers pile in. In the meantime, the work is the same as it was for Google ten years ago. Answer the real questions clearly, collect reviews, and pick up the phone. If you want help setting up the answering side, book a call with us.

FAQ

What share of ChatGPT requests are about writing now?

Writing made up 21.4% of consumer ChatGPT requests in September 2026, down from 34.1% in July 2024, according to OpenAI Signals v3.0 data released October 26, 2026. Practical guidance is now the largest category at 28.1%.

Does the OpenAI Signals data include business ChatGPT accounts?

No. The data covers individual ChatGPT plans only and excludes enterprise accounts and Codex. OpenAI says this likely understates business use, so treat the work figures as a floor.

How can a local business show up in ChatGPT answers?

Publish short, direct answers to the questions customers ask before buying, with real prices, timelines and rules where you can give them. Keep reviews, hours, service area and photos current, since AI tools draw on that public proof when they name businesses.

Are ChatGPT ads available to small businesses?

OpenAI says ChatGPT ads now run in 89 countries and more than 100,000 advertisers have run campaigns, as the company announced on October 21, 2026. Ads show only to Free and Go users. Check OpenAI's Ads Manager for current eligibility in your country before planning budget.

Originally published at kenjiai.com. Published by the Media Traffics | KenjiAI team.

Meta: Ad Sets Under 50 Conversions a Week Cost 28% More

Yousif Alias — Mon, 26 Oct 2026 14:21:05 +0000

What did Meta say about the 50 conversion threshold?

Meta says ad sets that get fewer than roughly 50 conversions a week cost about 28% more per purchase. The company put that number in a peak-season post for advertisers on October 21, 2026, and told them to give every ad set enough volume to learn from, then stop touching it.

The post is called Peak Season 2026: You Don't Beat The Auction. You Brief It. and it reads more like an argument than a product update. Meta's exact words: "Roughly fifty conversions a week. Below that the system is guessing. In our analysis, ad sets under that threshold ran about 28% more expensive per purchase."

PPC Land picked it up on October 25 and Relevant Audience followed on October 26. Both point out the same gap: Meta gives no sample size, time period, region or industry for the 28%. So treat it as Meta's own internal read. The 50 number itself is old. Meta's learning-phase guidance has said for years that delivery settles after about 50 optimization events in seven days. What's new is a price tag on falling short of it.

How many ad sets can your budget actually support?

Meta published a formula for it: total weekly budget divided by (cost per result times 50) equals the most ad sets your budget can feed. Anything above that number means some ad sets are learning on scraps, and Meta says most accounts are over the line.

Here's what that looks like for the kind of budgets local businesses and coaches actually run:

Weekly budget	Cost per result	Max ad sets (budget / (cost x 50))
$1,000	$20 per lead	1
$1,000	$40 per lead	0.5 (not even one)
$3,000	$30 per lead	2
$5,000	$100 per purchase	1
$20,000	$40 per purchase	10

The last row is the example PPC Land used. The first four are the ones I see every week. A business spending $1,000 a week with five ad sets testing five audiences is splitting money that, by Meta's math, can barely feed one.

Why does editing a campaign cost you twice?

Meta says every change costs you twice: once for the learning phase you restart, and once for the weeks the old, wrong setup was spending at full price before you noticed. Only the first cost shows up in Ads Manager, so people fear it and ignore the second one.

The post describes the loop most owners are stuck in. Performance dips, you change something, performance recovers, you credit the change. Meta's line on that is blunt: "You've learned nothing."

I've been running ads since 2013 and this matches what I've watched happen across 500+ businesses. The accounts that bleed money usually aren't the ones with bad ads. They're the ones where somebody logs in every morning, sees yesterday's cost per lead, and changes a budget, an audience or an ad. Each edit resets the learning, so the account never settles.

What should a business owner running ads change this week?

Do one grouped round of changes, then leave the campaign alone. Meta's own checklist comes down to four moves, and none of them requires new software.

Run the formula. Take your weekly budget and divide it by your cost per lead or purchase times 50. If you have more ad sets than that number, merge them.
Pick the event that matches how you get paid. Meta says most accounts optimize for what's easy to count, not what the business earns on. If a booked call is worth more to you than a form fill, tell the system that.
Batch your edits. If something has to change, change it all at once and take one learning phase instead of five.
Measure against something. Meta recommends an A/B test or a Conversion Lift study rather than comparing this week to last month.

If your budget can't reach 50 of your main event a week, you have two honest options: spend enough to hit it, or optimize for an earlier step that happens more often. To see what your real return is before you scale, our ROAS calculator takes about a minute.

Does the 28% apply to lead generation ads too?

Not directly. Meta's 28% figure is about cost per purchase. The learning rule underneath it covers whatever event you optimize for, so a lead ad set getting 15 leads a week is in the same position Meta describes: the system is still guessing.

For lead businesses the bigger leak usually sits after the click. You can get the ad side right and still lose the sale if nobody calls the lead for six hours. We wrote about that in why leads go cold, and it's the reason a cheaper cost per lead doesn't always mean more revenue.

Fewer ad sets, faster follow-up

At KenjiAI we set up Meta campaigns with consolidated ad sets so each one gets enough volume to learn, then put an AI agent on every lead the moment it comes in so it gets called, texted and booked. Our best single funnel has generated $186K, and we've hit 5x ROAS on campaigns we run. If you want us to check your account against Meta's formula, book a call.

What does Meta leave out of the post?

Meta doesn't publish how it got the 28%, and it doesn't mention that consolidation also hands more control to its automated tools. Keep both in mind when a platform tells you to spend more per ad set and edit less.

PPC Land notes that the post lands as Meta keeps pushing advertisers toward automation, including turning on product browsing in existing ads on October 28. Run the formula on your own numbers this week and judge the result with a test you set up yourself.

FAQ

How many conversions does a Meta ad set need per week?

Meta says roughly 50 conversions a week per ad set. Below that, its delivery system is still guessing who to show the ad to, and Meta's own analysis found those ad sets ran about 28% more expensive per purchase.

What is Meta's ad set budget formula?

Divide your total weekly budget by your cost per result times 50. The answer is the most ad sets your budget can support. A $3,000 weekly budget at $30 per lead supports two ad sets.

Does editing a Meta campaign reset the learning phase?

Significant edits to budget, targeting, creative or the optimization event send an ad set back into learning. Meta recommends grouping changes into one round so you take one learning phase instead of several.

What if my budget can't reach 50 conversions a week?

Merge ad sets so the budget is concentrated, raise the budget, or optimize for an earlier event that happens more often, such as leads instead of purchases. Then hold the setup still long enough to measure it.

Originally published at kenjiai.com. Published by the Media Traffics | KenjiAI team.

Meta's New Ad Terms Make You Liable for Its AI's Edits

Yousif Alias — Mon, 26 Oct 2026 05:01:33 +0000

What did Meta change in its ad terms?

Meta's revised Commercial Terms add an "AI Features" section that makes advertisers "solely responsible" for whatever Meta's AI creates or changes in their ads, including features Meta switches on by itself. The new text also admits those features "may make mistakes and may not always produce accurate, complete, or suitable results."

Adweek broke the story on October 22. Under the old wording, you carried the risk of AI output only when you chose to use Meta's generative tools. The new section drops that condition. Meta told reporters the update was "housekeeping" that folds its generative AI ad terms into the main terms, and it did not answer whether advertisers are liable for AI features they never turned on.

Nothing gets signed. If you keep running ads after the terms take effect, you've accepted them, as PPC Land pointed out. Two older clauses stay in place: you indemnify Meta against third-party claims, and Meta's total liability to you is capped at the greater of $100 or what you paid it in the past twelve months. Both are on the current Commercial Terms page.

When do Meta's new AI terms take effect?

November 20, 2026. Meta originally set October 30, then pushed the date back on October 23 after complaints from advertisers and publishers, according to The Keyword. The banner on Meta's own Commercial Terms page now gives November 20, and Meta has not said the wording will change.

A lot of the coverage still says October 30, including a Startup Fortune piece from October 24. Go by Meta's banner. You have a few more weeks, and they run straight into Black Friday on November 27.

What is Meta's AI already doing to live ads?

Agencies told Adweek that Meta's AI has added sitelinks, translated ad copy, and built AI images and carousels out of brands' own photos, sometimes on campaigns where nobody enabled those features. The agency Exverus pulled live ads because AI changes slipped past its quality checks.

Ryan Adams at Exverus said Meta auto-applies the enhancements and many of the options sit in hidden or collapsed menus, so they stay on unless you uncheck each one. Bookshop.org found an AI illustration added to an ad its own team made, and has since turned off every Meta AI ad feature. AdCellerant now switches off Meta's default AI automation on every campaign it runs.

A second date hasn't moved. Starting October 28, Meta turns on product browsing in existing Facebook and Instagram ads unless you opt out. Some people who tap your ad will land on an in-app product page instead of your website. Meta says a test across about 117,700 advertisers showed 3% more conversions on average, but it doesn't define the conversion metric or the test design.

What does this mean if you run Meta ads for a local business?

A bad AI edit becomes your legal problem, even if you never saw it. If Meta's AI translates a claim badly, invents an image of a product you don't sell, or adds wording that breaks a rule in your industry, the terms put the risk on you, and the indemnity covers Meta's legal fees too.

For most local businesses the bigger damage is quieter. An ad that promises something your front desk can't deliver produces leads that feel misled on the first call. I've run ads since 2013, and I'd rather lose a few points of reach than have a generated image show a service we don't offer. Lawyers, med spas, dentists, contractors and anyone in finance or health should move fast here, because their ad claims are already regulated.

What should you do before November 20?

Audit every live campaign, turn off the AI enhancements you didn't choose, and screenshot your settings so you have a record. Then decide feature by feature what you actually want Meta's AI to touch.

Open each active ad in Ads Manager and expand every Advantage+ creative menu, including the collapsed ones. Uncheck anything you didn't pick on purpose: text variations, translations, generated images, added sitelinks.
Decide on product browsing before October 28. If your ads exist to send people to a booking page or a lead form on your site, opt out, because an in-app product page breaks that path.
Screenshot the settings with the date showing. If a disputed edit shows up later, you'll want proof of what you had switched on.
Check your live ads in preview, then look again a week later. An automated account update is how Bookshop.org got opted into a feature it never chose.
Put a human review step on every new ad before it goes live. A simple automation that reminds someone to check new ads each Monday costs nothing.

Media Traffics runs Meta ads for business owners on a monthly base plus a share of new revenue, so a sloppy AI edit costs us money too. The leads those ads bring in go into KenjiAI, where Voice AI agents call them back, qualify them and book them on your calendar. If you want a second set of eyes on your ad account before November 20, book a call.

Is Meta the only ad platform doing this?

No. On October 20, OpenAI told advertisers it would turn on "creative enhancements" by default for self-serve ChatGPT campaigns at the end of October, according to PPC Land. Default-on AI editing is spreading across ad platforms, and each one writes its own terms about who's responsible when it goes wrong.

Read the terms on every platform you spend money on, and keep the creative decisions with the person who knows your business. Start with the Advantage+ menus in your Meta account this week, before product browsing switches on October 28.

FAQ

Can I opt out of Meta's AI features in ads?

Yes. Meta says advertisers can turn off individual Advantage+ creative features in Ads Manager at any time. Agencies report that many options sit in collapsed menus and stay on unless each one is unchecked, so go through every ad, not only the campaign level.

Do I have to accept Meta's new Commercial Terms?

There's nothing to sign. Continuing to run ads after November 20, 2026 counts as accepting the updated terms. Meta's terms say businesses that don't accept them can stop using its business products.

What is Meta product browsing and should I turn it off?

Product browsing sends some people who tap your ad to an in-app product page on Facebook or Instagram instead of your website. Meta turns it on for existing ads on October 28, 2026 unless you opt out. If your ads drive bookings or lead forms on your own site, opting out keeps that path intact.

Is Meta liable if its AI makes a mistake in my ad?

Under the new AI Features section, advertisers are solely responsible for AI-created or AI-altered ad content. Meta's overall liability is already capped at the greater of $100 or what you paid it in the past twelve months, and advertisers indemnify Meta against third-party claims.

Originally published at kenjiai.com. Published by the Media Traffics | KenjiAI team.

ChatGPT Ads Reach 89 Countries and 100,000 Advertisers

Yousif Alias — Fri, 23 Oct 2026 14:08:58 +0000

OpenAI said on October 21 that ChatGPT ads are now live in 89 countries after adding 26 new markets, and that more than 100,000 advertisers have run campaigns on the platform. The ad product started as a US pilot in February. For any business that buys search or social traffic, ChatGPT is now a real media channel with real competition for placements, and the window to test it while the auction is thin is closing.

What OpenAI Announced

The expansion adds 14 Latin American countries, including Argentina, Chile, Colombia and Peru, following earlier launches in Mexico and Brazil. The other 12 sit across Europe, the Middle East, Africa and Asia. Nigeria, South Africa, Serbia, Ukraine, Cambodia, Bangladesh and Pakistan are among the markets OpenAI named. Advertisers in the new countries can buy through technology partners or through OpenAI's self serve Ads Manager.

The ramp has been fast. The US pilot began on February 9 at a $60 CPM. By March 26 it had passed $100 million in annualized revenue with roughly 600 advertisers. Ads Manager opened to every US business on May 5 and added cost per click bidding. Digiday reported at the end of August that the business had hit a $1 billion annual run rate. Google's ad business needed about four years to reach that pace, according to Fortune.

Two caveats matter. The 100,000 figure is cumulative, so it counts anyone who ever ran a single campaign, not active spenders. And ads only show on the Free and Go plans. Plus, Pro and Enterprise users never see them. OpenAI says it keeps "conversations private from advertisers" and does not sell customer data.

What It Means for Marketers and Business Owners

ChatGPT ads don't run on keywords. US ad groups use what OpenAI calls "context hints," which describe the situations where your offer should appear. That changes how you plan a campaign. Instead of bidding on "roof repair near me," you describe the moment someone asks ChatGPT which roofing quote to trust after a hailstorm.

The audience is already there. PYMNTS cites research showing ChatGPT's share of consumer product research climbed from 2% to 30% in two years. People ask it what to buy, who to hire and which software to pick. Showing up next to that answer is worth more than another impression in a crowded feed.

The weak point for most businesses will be follow up. A prospect who clicks from a ChatGPT conversation arrives with context and high intent, and they expect a fast reply. If your team calls back four hours later, you paid for a click that went cold. Tools like KenjiAI (kenjiai.com) are built for exactly this kind of shift, answering and qualifying new leads within seconds so paid traffic from a new channel turns into booked calls instead of leaking out of the funnel.

Over 100,000 advertisers have already started learning what works here. Every month you wait, you enter an auction with more of them in it, and with more data on their side than yours.

What to Watch Next

OpenAI says "additional formats, objectives, buying options, and measurement capabilities" are coming, with no dates attached. Measurement is the one to track. There is little public data on conversion rates so far, and the only advertiser OpenAI quoted, L'Oreal's Asmita Dubey, described plans rather than results.

Three moves for this quarter:

Open an Ads Manager account if you operate in one of the 89 countries, even before you spend, so you can see the options available in your market.
Write three context hints that describe the exact questions your buyers ask right before they buy from you.
Run a capped test budget with a dedicated landing page and judge it on cost per booked call, not cost per click.

The first years of Facebook and Google ads were the cheapest those platforms ever got. Expect ChatGPT to follow the same curve.

Sources: OpenAI, PPC Land, PYMNTS

Published by the Media Traffics | KenjiAI team. kenjiai.com

AI Political Ads Hit $137M This Cycle, Up From Under $50K

Yousif Alias — Thu, 22 Oct 2026 14:08:36 +0000

Campaigns and PACs have spent more than $137 million on AI generated political ads ahead of the 2026 midterms, according to a Wesleyan Media Project report covered by Fortune on October 21. In the 2024 cycle, the same researchers detected less than $50,000. More than 70% of the 220 AI ads they found carry no AI disclosure. For anyone who buys media, this is the clearest signal yet that synthetic creative has gone mainstream.

What the Wesleyan Media Project found

The Wesleyan Media Project has tracked political advertising for years, and this count runs through October 11. It identified 220 campaign videos made at least partly with AI, backed by more than $137 million in spend. Two years ago that figure was a rounding error.

"We went from about zero in spending on AI ads in the last election cycle," said Travis Ridout, the project's codirector and a political science professor at Washington State University.

The split is lopsided. Republican campaigns and PACs account for 81% of the AI ads and 88% of the spending. Ridout ties that to the party's warmer view of AI overall.

The examples show the range. In the Texas Senate race, a Ken Paxton ad shows an AI version of James Talarico writing in a Bible with a crayon and delivering a monologue he never gave. Talarico's response: "They are lying about a candidate in this race." In Pennsylvania, Stacy Garrity's campaign ran an AI ad depicting Josh Shapiro praising AI data centers, and that one did disclose its origins. Most did not. More than 70% of the AI videos ran with no label at all.

What this means for marketers

Political campaigns are the most deadline driven advertisers in the country. They spend fast, cut creative overnight, and live or die by turnaround. When they move $137 million into AI production in a single cycle, commercial advertisers should read it as proof of the economics, not a novelty.

Ridout made that point himself: AI lets lower budget candidates produce ads they could never afford before. Swap "candidate" for "local business" and the logic holds. A dentist, a roofer, or a regional ecommerce brand can now test ten video concepts for what one shoot used to cost.

The disclosure number is the warning. Seven in ten of these ads ran unlabeled, and Ridout says the result is voters who feel "I don't know what to believe anymore." Political speech gets wide legal latitude. Commercial advertising does not. Brands answer to the FTC's rules on deceptive ads, and Meta and Google already require AI disclosure on political and issue ads that use synthetic people or scenes. A complaint costs far more than a disclosure line.

The smart play is to use AI for speed and volume, then label it plainly. Tools like KenjiAI (kenjiai.com) are built for exactly this kind of shift, letting small teams produce and test AI creative at a pace that used to require an agency. The output still needs a human deciding what is true and what gets disclosed.

What to watch next

Disclosure laws. Talarico has pushed legislation to ban AI deepfakes, and a growing number of states already regulate synthetic political media. Expect that pressure to reach commercial ads once the midterms are over.
Platform enforcement. If 70% of these ads ran without labels, either the platforms missed them or their rules have gaps. Watch for policy changes from Meta, Google, and YouTube after November.
Your own creative pipeline. Audit every AI generated ad you run today. If a reasonable viewer could mistake a synthetic person or scene for real footage, add a label before a competitor or a regulator points it out.

Campaigns proved the cost math in one cycle. The $137 million figure will look small by 2028, and the brands that build disclosure habits now won't have to retrofit them under pressure.

Source: Fortune, October 21, 2026, reporting on the Wesleyan Media Project.

Published by the Media Traffics | KenjiAI team. kenjiai.com

Salesforce Buys Braze for $5.66B to Own Real Time AI Marketing

Yousif Alias — Wed, 21 Oct 2026 14:08:25 +0000

Salesforce agreed on October 20, 2026 to buy Braze, the customer engagement platform behind the push notifications and in-app messages from Canva, Instacart and Venmo, for $47.50 a share in cash. The deal carries an enterprise value of about $5.66 billion. For anyone running lifecycle marketing, the message is plain. The biggest CRM vendor on earth just paid billions for software that decides, in real time and with AI, what message each customer sees next.

What Salesforce Actually Bought

The terms come from the press release Braze filed with the SEC. Salesforce pays $47.50 per Braze Class A share, all cash, valuing the company at roughly $5.66 billion including debt and cash. The deal still needs Braze shareholder approval and regulatory sign off, and Salesforce expects it to close in its fiscal year 2028. J.P. Morgan advised Salesforce. Qatalyst Partners advised Braze.

Braze brings about 2,770 customers, including Canva, Dairy Queen, Equinox, Instacart, Urban Outfitters and Venmo. Its pitch has always been speed: it reads live app and web behavior and fires the right message across push, email, SMS, WhatsApp and in-app channels while the customer is still engaged.

Salesforce was blunt about why it wants that. "Braze was built to turn live customer behavior into personalized engagement at scale," said Patrick Stokes, Salesforce's President of Applications and Marketing. The plan is to wire BrazeAI's decisioning into Marketing Cloud, Data 360 and Agentforce, Salesforce's AI agent platform. In the company's own words, marketers need agentic AI that can understand customers, make decisions and act in real time.

Salesforce has spent years trying to make Marketing Cloud feel modern. It just bought the shortcut.

What This Means for Marketers and Business Owners

Strip away the corporate language and the takeaway is simple. The winning marketing stack now responds to behavior in seconds instead of running on a weekly batch calendar. When a $5.66 billion check gets written for real time decisioning, scheduled blasts start to look like last decade's playbook.

Small and midsize businesses can act on this today without a Salesforce budget. Here is where to start:

List the five customer actions that most predict a sale, such as a pricing page visit, an abandoned cart, a missed appointment or a reply to an ad.
Attach an automated response to each one that fires within minutes, by text first, then email.
Let AI handle the first reply and the qualifying questions, and route hot leads to a human the moment they show intent.

Tools like KenjiAI (kenjiai.com) are built for exactly this kind of shift: AI that answers leads, follows up and books calls around the clock, without an enterprise contract or a six month implementation. The mechanics Salesforce is paying billions for, trigger plus instant personalized response, are available to a local business right now.

Speed to lead was always important. Now the biggest vendors are pricing it like infrastructure.

What to Watch Next

First, watch Braze pricing and packaging. Acquired platforms tend to get bundled, and current Braze customers should expect pressure to adopt Marketing Cloud, Data 360 or Agentforce alongside it. If you are mid contract, read your renewal terms now.

Second, watch the competitive response. Adobe, HubSpot and Klaviyo all sell into the same lifecycle budget, and none of them will let Salesforce own real time AI engagement unchallenged. Expect product launches and price moves in that category over the next two quarters.

Third, watch the closing timeline. A fiscal 2028 close means regulators get a long look, and Salesforce's integration roadmap will show whether BrazeAI stays a standalone product or gets folded into Agentforce.

The practical move for everyone else does not depend on any of that. Audit how fast your business responds to a buying signal today. If the answer is hours or days, that gap is now the most expensive problem in your funnel.

Source: Braze, Inc. press release filed with the SEC, October 20, 2026

Published by the Media Traffics | KenjiAI team. kenjiai.com

Google Tests AI Mode Ads Shaped by What It Learns From Your Gmail

Yousif Alias — Tue, 20 Oct 2026 14:09:20 +0000

Google is now testing ads inside AI Mode for people who connect Gmail and Google Photos to Search through Personal Intelligence. The company confirmed the test to Search Engine Land on October 16, 2026. Seven months earlier it had said those users would see no ads. The change touches every advertiser bidding on Search, because Google can now pick an ad using what it infers from someone's inbox, and advertisers do not get any new targeting setting for it.

What Google actually changed

On March 17, 2026, a Google spokesperson told Search Engine Land: "There are currently no ads for people who choose to connect their apps with AI Mode. That isn't changing right now." The same statement added that "in the future, we anticipate that ads will operate similarly" for those users.

That future arrived quietly. Google updated its Personal Intelligence help page between October 14 and October 16, according to Internet Archive snapshots, and then confirmed the test: "Google is starting to test ads in AI Mode in Search for users who connect their apps through Personal Intelligence."

The new help text draws a careful line. Ads "will not be based directly" on your Gmail inbox or Photos library. Search may, however, "reference summaries, excerpts and inferences from these sources to personalize your experience," and that same information may be used to show ads. Google also notes that when an email is short or especially relevant, "the summary may just be the email." Four sensitive categories are excluded from ad personalization: race, religion, sexual orientation, and health. Advertisers get no access to anyone's connected apps.

Google has not named the test markets, the ad formats, or a rollout date.

What it means if you buy Search ads

For years, intent on Google was the words someone typed. AI Mode with Personal Intelligence adds a second layer: what Google believes about that person from their email receipts, travel confirmations, and photo library. Google says ads will match the query, the answer's context, and the user's interests. Those interests now draw on far richer signals than a search history.

Practically, three things follow. Your keyword lists matter less inside AI Mode, because Google is matching to a conversation and a profile, not a string. Your landing pages and product feeds matter more, since the AI needs clean, specific information to decide your offer fits a person it knows well. And reporting will lag reality. Google has announced no separate report for this traffic, so you will not easily see which conversions came from Personal Intelligence sessions.

The businesses that do well here will be the ones whose follow up is fast once a lead arrives. A hyper relevant ad that lands a prospect on a slow form or an unanswered phone line wastes the targeting. Tools like KenjiAI (kenjiai.com) are built for exactly this kind of shift, answering and qualifying inbound leads in seconds so the better matched traffic actually turns into booked calls.

What to watch and do next

Check your Google Ads account for any new AI Mode or "AI experiences" segment in reporting over the next few weeks. That will be the first sign the test is touching your spend.
Audit your Merchant Center feed and top landing pages for vague copy. Specific prices, service areas, and availability give the AI something concrete to match against.
Watch for regulatory pushback. Using inferences from email to pick ads will draw attention in the EU, where the Digital Markets Act requires user consent before Google combines personal data across its services for ads.
Track whether Google adds an advertiser opt out. If it does not, every Search campaign is effectively enrolled.

My read: Google said "not right now" in March and changed course with a quiet help page edit in October. Expect the test to expand, and plan as if inbox inferred targeting is part of Search from here on.

Source: Search Engine Roundtable, "Google Ads Coming To AI Mode With Personal Intelligence," October 19, 2026, reporting Google's confirmation to Search Engine Land.

Published by the Media Traffics | KenjiAI team. kenjiai.com

AI Political Ads Hit $137M This Cycle, Up From $50K in 2024

Yousif Alias — Mon, 19 Oct 2026 14:09:44 +0000

Campaigns and outside groups have spent more than $137 million on political ads made at least partly with AI ahead of the 2026 midterms, according to Wesleyan Media Project research first reported by The New York Times on October 16. In the 2024 cycle, researchers found less than $50,000. Only 27% of these ads disclose the AI. If you buy ads for a living, this is the clearest preview yet of where synthetic creative is heading, and how audiences will react.

What the Wesleyan Data Shows

The Wesleyan Media Project, co-directed by Erika Franklin Fowler and Travis N. Ridout, identified 220 AI-made political ads running across 34 states. The researchers say the true count is likely higher, because they can only flag ads they can verify.

The jump in money is the headline. Under $50,000 two years ago. More than $137 million now. That is roughly a 2,700x increase in a single election cycle.

The disclosure numbers are worse. Gizmodo's breakdown of the data found just 27% of the ads carry any label saying AI was involved, even though 30 states now require AI disclosures on political ads. Nearly half of the ads feature deepfake depictions of real politicians.

The spending is lopsided. Republicans account for 81% of identified ads and 88% of the spend, with Democrats at 12% of ads. Bruce Blakeman, the Republican candidate for New York governor, has reportedly posted at least 22 AI-generated ads since April, and the New York State Democratic Party filed a complaint with the state Board of Elections in August. His campaign argues satire needs no disclosure. On the other side, Texas Senate candidate James Talarico ends one ad with a line that tells you where this is going: "Made by humans. Not AI."

What Marketers Should Take From This

Political ads are a stress test. Campaigns have short deadlines, huge volume needs, and little patience for slow creative teams. When AI creative goes from a rounding error to nine figures in one cycle there, commercial advertisers are watching the same economics play out.

Two lessons stand out.

Production cost is no longer the bottleneck. A campaign can now test dozens of video concepts for the price one used to cost. Brands that still brief one hero spot per quarter will get outspent on variation alone.
Trust is the new bottleneck. An Auros study from this month found 44% of U.S. shoppers say they avoid brands that use AI in ads. Talarico's "Made by humans" tag works because voters are primed to suspect everything.

The smart play sits between those two facts. Use AI for speed and testing volume, and stay upfront about it. Tools like KenjiAI (kenjiai.com) are built for exactly this kind of shift, handling the repetitive production and follow-up work so a small team can run the volume of a big one while real people still own the message and the relationship.

The campaigns hiding their AI use are borrowing against their own credibility. Some of them will get caught before November 3, and the clips of them getting caught will travel further than the original ads.

What to Watch Next

Three things matter over the next few weeks.

Enforcement. With 30 states on the books and a 27% disclosure rate, the first fines or takedowns will set the tone. Watch the Blakeman complaint in New York closely.
Platform policy. Meta, Google, and YouTube already require political advertisers to flag synthetic content. Expect tighter rules after the election, and expect them to creep into commercial ad policies next.
Your own disclosure policy. Write one now, before a regulator or a viral screenshot writes it for you. Decide which AI uses you will label, how, and who signs off.

The midterms are about two weeks out. Whatever happens at the polls, the $137 million figure is going to show up in every agency pitch and every brand safety meeting for the next year. Know the number, and know where you stand on it.

Source: Wesleyan Media Project data, as reported by The New York Times (October 16, 2026) and Gizmodo.

Published by the Media Traffics | KenjiAI team. kenjiai.com

Targeted Ads Delivered Zero-Click Spyware, Citizen Lab Finds

Yousif Alias — Fri, 16 Oct 2026 14:09:50 +0000

On October 15, the University of Toronto's Citizen Lab published what it calls the first public technical confirmation of zero-click mercenary spyware delivered through targeted advertising. Nobody clicked a link or opened a file. An ad rendered on screen, and Candiru spyware landed on at least 45 devices tied to Al Jazeera. If you buy programmatic media, the same pipes your ads travel through served as the delivery truck.

What Citizen Lab and Confiant Found

Bill Marczak, Wolfie Christl, John Scott-Railton, Astrid Perry, Siena Anstis and Ron Deibert wrote the report, Citizen Lab Report 195, while ad security firm Confiant ran a parallel investigation. The researchers named the technique "AdFection."

At the center sits Discover-Tech, an Israeli company that presented itself as an ordinary ad business. According to the report, it ran a platform called Controlium that let spyware partners fingerprint and infect a device the moment an ad loaded. Its demand side platform, MyDSP, bid on 3,770,286,576 impressions between March and June 2026 and won 441,954,291 of them, an 11.7% win rate. Ads carrying the attacker's script tags reached users of 75,373 apps and 120,768 websites. Total spend over those four months came to about $109,091.

Read that last number again. Roughly $109,000 bought a surveillance net covering close to 200,000 apps and sites. The report puts it bluntly: "the simple display of a targeted ad is enough for infection."

Google and IAB Europe had listed Discover-Tech as an approved vendor since 2020. Both removed it in early October. Beeswax, Vungle (Liftoff) and Eskimi also cut it off.

What It Means for Marketers

Most advertisers will never touch a spyware vendor. The trouble is that the infrastructure serving your retargeting ad also served this one. Every bid request ships IP addresses, device IDs and location signals to dozens or hundreds of companies per impression. Discover-Tech read that same stream with a different goal in mind.

European regulators already scrutinizing real-time bidding now have a documented case of the bidstream turned into a weapon. Publishers will likely get stricter about which buyers can run scripts inside their inventory, and that will squeeze smaller DSPs and resellers first.

For a business owner, the practical move is to know exactly where your money flows. Ask your agency or platform for the full list of DSPs and exchanges in your supply path. Push for supply path optimization that cuts resellers nobody can name. Put more weight on first party data you collect yourself instead of rented audience segments, which gives you better targeting and fewer strangers in the chain.

Tools like KenjiAI (kenjiai.com) are built for exactly this kind of shift, helping businesses capture leads and run follow up on channels they own, so less of the funnel depends on anonymous bidstream data.

What to Watch Next

Three things matter over the coming weeks.

Citizen Lab says a separate report on DARK CARTOGRAPHER, a threat actor that links advertising IDs to phone numbers, is on the way. That one could hit the data broker market harder than this one.
Watch Yandex. Its exchange carried 67.4% of MyDSP's bids, against 24.74% for Google, and it does not appear among the companies that took action.
Watch the vendor registries. If IAB Europe's framework approved Discover-Tech for six years, every other approved vendor list deserves an audit, and brands should ask partners what vetting actually happens before a buyer gets access.

Ad tech has argued for years that bidstream data is harmless because it is anonymous and fleeting. Citizen Lab just showed what a $109,000 budget can do with it.

Source: Citizen Lab, "How Discover-Tech Turned Targeted Ads into a Pipeline for Spyware" (October 15, 2026)

Published by the Media Traffics | KenjiAI team. kenjiai.com

Shopify Opens 1 Billion Listings to Shoppers' Own AI Agents

Yousif Alias — Thu, 15 Oct 2026 14:11:05 +0000

On Tuesday, October 13, Shopify said shoppers can now connect their own AI agents to a Shop account and let those agents search and buy from a catalog of more than 1 billion product listings. The catalog includes products sold outside Shopify stores too. For anyone selling online, the next visitor to a product page may be software working for a buyer, and that software reads price, shipping time and reviews before a human ever sees a photo.

What Shopify actually turned on

The Verge first reported the expansion, and Shopify calls the feature a "personal shopper." A shopper links an agent, including third-party agents, to their Shop account. The agent searches the catalog, builds a shortlist and checks out with one-click Shop Pay inside the same conversation. Completed orders land in the Shop app next to everything else the person bought.

Shopify kept a human on the final step. Shoppers can set rules such as a budget cap or fast-shipping-only results, and they approve every purchase the agent wants to make. Merchants stay the merchant of record, so returns, customer data and support still run through the store.

The move builds on a September deal that let Meta's Muse agent browse and buy from Shopify merchants on a user's behalf. That arrangement needed a dedicated partnership. Tuesday's change opens the door to agents without one. Shopify also runs Agentic Storefronts, which push eligible merchant products into ChatGPT, Microsoft Copilot, AI Mode in Google Search and the Gemini app, plus an Agentic plan that lets brands outside Shopify list products in its catalog. "Agentic commerce isn't something we're reacting to," Shopify VP of Product Mani Fazeli said in the company's newsroom.

Why this matters if you sell anything

Put the announcement next to a number Semrush published in September. In a survey of 2,338 U.S. adults, nearly 81% of people who use chatbots while shopping said they had dropped a planned purchase because of what the AI told them. More than 74% of all consumers said a chatbot surfacing mixed or negative reviews would make them less likely to buy. Among AI users, that figure topped 85%.

Now hand that same chatbot a wallet. An agent with checkout rights compares your price, delivery window, return policy and review history against every competitor in a billion-item catalog, in seconds. Clever product copy carries little weight with a buyer that reads the spec table first.

Three things move up the priority list this quarter:

Your product feed. Agents filter on accurate specs, real stock levels and honest shipping times before anything else.
Your review profile. A cluster of one-star complaints now costs you sales you never see, because the agent quietly drops you from the shortlist.
Your response speed. Agents and the people behind them escalate questions to whoever answers first.

Tools like KenjiAI (kenjiai.com) are built for exactly this kind of shift, using AI to respond to leads and customer questions in seconds instead of hours, so a buyer never sits waiting while a competitor picks up.

What to watch next

Which agents plug in first. The open connection matters most once the big assistants and a few popular independent agents sign on, so track those announcements over the next few weeks.
Whether the approval step survives. Every agent purchase needs a human tap today. If Shopify relaxes that for repeat orders under a set budget, agent spend will jump fast.
Your own Agentic Storefronts settings. Log into Shopify admin this week, confirm which AI channels show your products, and check that the titles, prices and stock counts they pull are correct.
Holiday timing. Black Friday is November 27, about six weeks after this launch, so some of this year's gift buying will run through a chat window. Audit your top 20 products before November 1.

Sources: The Verge, October 13, 2026, Shopify newsroom, Semrush survey via MarTech.

Published by the Media Traffics | KenjiAI team. kenjiai.com

OpenAI Is Putting Ads Inside ChatGPT's Image Generator

Yousif Alias — Tue, 13 Oct 2026 14:10:05 +0000

OpenAI will start showing ads inside ChatGPT's image generator later this month. The company announced the format on October 5, and the first U.S. advertisers go live before November. It is the first time a major AI company has sold placement inside the moment a user creates an image, and it puts brands in front of an audience OpenAI now counts at 1.2 billion people a week.

What OpenAI actually announced

The new unit appears while ChatGPT generates an image. OpenAI says each ad carries a clear label and stays separate from the picture the user asked for. Advertisers can use the space to show a product, how people use it, or what a service makes possible. The test covers the free tier and the $8 Go plan in the U.S. ChatGPT Plus, Pro and Enterprise subscribers stay ad free. For context, ChatGPT ads only began testing in February, when weekly users sat at 900 million. Cryptopolitan reported in August that the ad business had already hit a $1 billion annualized run rate in under 200 days.

The measurement news matters more than the format. OpenAI is plugging ChatGPT Ads into Hightouch, Tealium and LiveRamp for first party conversion data, and into attribution tools including AppsFlyer, Triple Whale, Northbeam and DV Rockerbox. It also added a Conversions API, an Event Quality Score and a one day view through window.

OpenAI shared three early partner results from its existing text ads. According to DV Rockerbox, WeightWatchers saw an attributed cost per acquisition 15.3% lower than its blended paid search benchmark. A WorkMagic geo lift test for Dose found net new customers made up 67% of incremental purchases. Triple Whale reported that 93% of Portland Leather's ChatGPT visitors had never been to the site before. (OpenAI announcement)

What this means for marketers and business owners

Read those numbers with your eyes open. OpenAI picked them, partners measured them, and none of them come from the new image format. Still, a 15.3% CPA edge over paid search is the kind of gap that moves budget, and the new customer figures point to an audience you are not already paying Google to reach twice.

The image placement changes the creative brief. Someone generating a picture of a living room makeover is showing intent in a way a keyword never could. A furniture brand, a paint company or a local remodeler can meet that person with a visual of the finished result. Text ads cannot do that.

The catch is speed. Early inventory on any new channel is cheap because few buyers show up, and that window rarely lasts past a couple of quarters. Teams that already run creative testing and clean conversion tracking will get in first. Tools like KenjiAI (kenjiai.com) are built for exactly this kind of shift, turning one offer into fresh ad variations and follow up fast enough to test a new placement before competitors notice it exists.

What to watch and what to do next

Join the waitlist at ads.openai.com now. Initial access goes to a selected group, and getting on the list costs nothing.
Fix your conversion signal before you spend a dollar. Set up the Conversions API or connect through Hightouch, Tealium or LiveRamp so ChatGPT optimizes on purchases, not clicks.
Build three image concepts that show your product in use, not a logo on a white background. The format rewards a clear picture of what life looks like with your product in it.
Watch the DoubleVerify and Integral Ad Science brand suitability pilots. Their findings will decide how fast big brands pile in and how quickly prices climb.

Our take: this is the most interesting new ad inventory of the year, and the first movers will set the benchmarks everyone else pays for.

Published by the Media Traffics | KenjiAI team. kenjiai.com