Reddit tests scale, LinkedIn automates: the July 2 brief

Test budget got a cleaner price tag today. Reddit made proper ad experiments available to everyone, LinkedIn pushed more ad-writing work onto its AI, Google tested AI text inside paid search results, and HubSpot bought its way closer to knowing who is ready to buy. The platforms want to make more of your decisions for you. Your profit depends on which decisions you still check yourself.

Reddit puts a floor under ad testing

Reddit made Split Testing generally available on July 1. It lets advertisers run controlled experiments (show version A to half the audience, version B to the other half, and let the data pick the winner) inside its Ads Manager, with the winner called at 65% statistical confidence, according to Reddit for Business. The catch matters: Reddit says the feature has a $1,000-a-day minimum spend.

That makes this a cost-cutting tool only for advertisers already spending enough to learn cleanly. Reddit says beta data found 4 out of 5 split tests identified a version that returned more revenue per ad dollar, with ready-made templates for its automated targeting and budget features.

The money angle is waste avoidance. Without a controlled test, an owner compares two date ranges, two sets of ads, and a dozen outside variables pretending to be a decision. Change one thing at a time and the answer is real.

Smaller accounts should not force the spend just to use the feature. Put the same one-change-at-a-time discipline into cheaper channels first. Our rule from the Google Ads cleanup still applies: isolate the variable before trusting the winner.

LinkedIn shifts B2B ad work into the machine

LinkedIn introduced new ad-creation tools on July 1, including a brand kit, AI ad drafting, ad personalization, AI-generated ad variations and a mix-and-match format, according to the LinkedIn Marketing Blog. The mix-and-match tool combines the images, videos and copy you supply, then shifts budget toward the combinations that perform as the results come in.

The money lever is output. Companies selling to other businesses often under-test their ads because every headline and variation takes someone's time. LinkedIn is trying to compress that cost. It says the mix-and-match approach generated roughly 7% more ad combinations to test.

The failure mode is false confidence. More variations do not mean more learning if every variation says the same thing with a different adjective. You still own the offer, the promise, and the page the click lands on.

If you advertise on LinkedIn, separate message testing from format testing. Group variations by angle: the problem, the outcome, the proof, the deadline. If one angle wins, feed that lesson back into your landing page, not just the next ad.

Google tests AI summaries inside sponsored results

Google is testing AI-generated summaries below the text of paid search ads, Search Engine Roundtable reported on July 1 after screenshots surfaced on X from Darcy Burk. The test unit included a disclaimer that Google's AI responses are generated independently and can make mistakes, according to Search Engine Roundtable.

This one hits the share of clicks that turn into customers. Search ads are expensive because the person clicking usually already wants the thing. If Google inserts its own machine-written comparison or caution inside your paid listing, the ad is no longer just your words. It becomes your words, Google's summary, and a trust warning in the same box.

That can help or hurt depending on what you sell. For services where buyers shop on price, an AI note that reframes cost or risk can pull clicks away before your page gets a chance. For high-trust purchases, it can filter out people who would have left anyway. Either way, your dashboard can misread the change as your ads wearing out when it was the ad box itself that changed.

Watch searches for your own name separately from generic searches, and mark any click or sales change around June 30 and July 1 in your records. This connects to AI Overviews and lead flow: Google keeps inserting more of its own interpretation between the searcher and your site.

HubSpot buys closer to the buying signal

HubSpot's acquisition of Warmly, covered by MarTech on July 1, points to the next battleground in customer databases: identifying which individual person is showing signs of being ready to buy, not just which company visited your website. Warmly is known for putting names to anonymous website visitors and using AI agents to help sales teams reach out.

The money lever is speed. Knowing a company visited your site is useful but slow: someone has to go research who to call. Knowing the person creates an immediate, routed follow-up. That shrinks the gap between "someone is looking" and "sales has a reason to reach out."

The risk is chasing weak signals. Not every website visit is a future sale. If AI agents turn every anonymous hit into a task, your sales system looks busy while your people chase noise. Better: only route the visits that combine a good-fit company, a high-intent page like pricing, and recency. Everything else stays in the slow-drip follow-up pile.

This is why customer databases keep moving beyond contact storage and toward follow-up mechanics. The same budget logic runs through our CRM guide for agencies: the tool matters less than whether your lead, booking and follow-up data sit close enough together to act on quickly.

What to watch

Cloudflare's new default policy of blocking AI crawlers could make your own email list and direct visitors more valuable, if more publishers start restricting unpaid scraping.

Google AdSense adding AI-written content to its ad units matters for anyone earning from website ads, because machine-written answers can change how the revenue splits on information pages.

TikTok's Agentic Hub is still early, but if it turns creator and shopping workflows into campaigns run by AI agents, the next fight in paid social is over who controls the brief.