19 Updates - Reels Amazon Links, ChatGPT Ads, AI Shopping
Happy Wednesday.
- Today I learned that sharks are older than trees. Well, sort of. Modern sharks showed up around 195 million years ago, long after the first trees at about 385 million. But count their spiny, shark-like relatives from 400 million years ago, and sharks win. It all depends on how you define a shark, which might make this the oldest attribution debate on Earth.
- ChatGPT ad spend more than doubled, Instagram Reels can now sell straight from the video, and shoppers still will not hand their card to an AI agent.
- Also today: OpenAI is testing Google-style layouts in ChatGPT, Tom Capper from Moz lists four prompt tracking mistakes, and LinkedIn is about to talk money.
Here is everything I am covering today.
SOCIAL MEDIA
Creators Are Turning Into Checkouts: Reels Get Amazon Links
In Glengarry Glen Ross, coffee was for closers only.
Now creators can close a sale straight from the Reel. Pay them like closers.
Reels with a register. Instagram now lets US creators earn Amazon's commission by tagging Amazon products in Reels, up to 30 each, and in Feed posts, up to 20 each.
Meta says these links get about twice the clicks of bio links. Sell on Amazon? Creators now earn on your products, even without a deal.
The long tail gets paid. Twitch says its brand-matching tool gets four times as many streamers sponsored as last year, after it opened sponsorships to smaller streamers in August.
Payouts only doubled: more deals, smaller checks.
Twitch ads can end at a checkout, too. Twitch's shoppable ads let viewers add an ad's products to their Amazon cart.
And remember that YouTube also went big on Amazon affiliate programs in September.
Someone has to run the shop. A new Net Influencer hiring report puts TikTok Shop roles at 2.52 times their 2023 share of marketing hires.
68% of those hires now land outside TikTok and ByteDance, in roles like affiliate manager and live host.
More signals of creator commerce? Unilever now works with over 500,000 creators, up from 75,000 in 2024, per its CMO. Nobody hand-negotiates that many flat fees. Commissions scale.
Hauls fall out of fashion. Haul videos, where a creator shows off a pile of products, halved to 1.2% of sponsored posts this year, per Traackr.
CreatorIQ says engagement per sponsored haul also fell about a third, and brands now favor pricier story-led videos about one product.
My take: split how you pay this holiday season. Offer commission-led deals to the long tail through TikTok Shop or Amazon affiliate links on Instagram, and keep flat fees for a few named creators making story-led videos, not hauls. The money still shows up, just after the sales do.
CHATGPT
US Spend on ChatGPT Ads More Than Doubles, but Most Chats Don't Look Like Shopping
ChatGPT's users prefer to always be chatting.
The checks keep clearing anyway. AdClarity, which tracks ad spend, estimates US advertisers spent $70.7 million on ChatGPT ads in October.
That is more than double April's $31.1 million. Across 24 countries, October hit $110.6 million.
Developer software takes about 25%, so software marketers have the most company.
Are the buyers there, though? A SparkToro analysis of ChatGPT chats from a US panel found only 6.9% showed purchase intent.
Most answers (87.9%) named no brand. The panel skews toward mobile, light users, so check your own ChatGPT referral and ad data before moving budget.
Starting to look familiar? OpenAI is testing Google-style layouts in ChatGPT:
- Web-only search results
- A short AI answer with a "See more" button, like Google's AI Overviews
If they stick, ChatGPT becomes a results page you can rank on.
Core updates, minus the warning. After OpenAI launched GPT-5.6 on July 9, AP, TIME and three other newsrooms lost 50% to 90% of ChatGPT citations within a week.
A possible signal that each new ChatGPT model reshuffles the sites it cites. Also, Google rankings did not follow this switch, so track citations per model, and mark OpenAI's launch dates.
Awkward timing. Three days after ChatGPT Wallet let agents pay for shoppers, California's attorney general sued OpenAI, calling its agent testing a public nuisance.
The complaint says OpenAI's test agents hacked the AI platform Hugging Face in July.
It says OpenAI admitted to over 80 loss-of-control incidents in testing, but publicly reported only 27. OpenAI says most were low severity.
OpenAI still wants merchants to wave its agents past bot and fraud filters.
My take: until more chats start closing, let your own data set the budget. The spend is real and growing fast, but 6.9% purchase intent means ChatGPT is still a conversation first and a checkout second.
AIO
Four Prompt Tracking Mistakes That Skew Your AI Visibility Numbers
Say ChatGPT recommends your product, then cites a review site instead of your own. Loss?
Not to Tom Capper, head of Search Science at Moz. He lists four prompt tracking mistakes that feed doubts about measuring AI visibility.
Mistake #1: Wanting the footnote more than the shout-out. The mention beats the citation. If anything, a third party vouching for you carries more authority than your own site. Count sub-brands too, like iPhone for Apple.
These answers rarely send clicks, so treat them as awareness. Do not like how an answer describes you? Pitch the site it cites.
One more wrinkle: ChatGPT's new Business Hub shows verified businesses their mentions. For online businesses, though, it only counts a mention when ChatGPT references the website, so it works more like a citation count. Keep tracking name-drops yourself.
Mistake #2: Mentioned first, outnumbered anyway. Coming first in an answer feels like ranking #1. But a rival brand can still show up more often in it.
Across many prompts, start with the share of answers that mention you at all. Then check how they describe you, down to the adjectives.
Mistake #3: Tracking thousands of keywords, but 50 prompts. Prompts run longer tail, and answers vary more by the day. So big brands need at least as many prompts as top-of-funnel keywords.
Fifty can be enough for a local or niche business, which tracks fewer keywords anyway.
The fix: rerun each prompt dozens of times and report how often you appear. SparkToro and Gumshoe put the odds of getting the same brand list twice at under one in 100.
Mistake #4: Tracking prompts nobody types. Hardly anyone opens ChatGPT, types "laptops" and hits Enter, so do not paste your keyword list into the tracker.
About 15% of Google's daily searches are brand new. For prompts, Tom guesses 90-something percent, most with a volume of roughly one. So pick your key topics and write realistic prompts, starting from Google's People Also Ask questions.
My take: once you fix those four, get creative. Add personas, locations, languages or qualities like "cheapest" to prompts, and ask what rivals do better. Use the same tracker. Just ask better questions.
DATA STORIES
Retailers Are Ready for AI Shopping Agents, but Few Shoppers Are
Retailers are setting the table for AI shopping agents, but very few shoppers plan to show up for dinner.
Talkdesk surveyed US retail leaders and holiday shoppers, and the two groups are far apart.
On the shopper side, the trust gap is wide. AI shopping agents are tools that browse and buy products on a shopper's behalf. 89% of retailers say they are preparing for them.
Only 15% of shoppers would let an agent buy with their payment details, and only if they can review the order first. Without that review step, the number drops to 10%.
That is roughly a 6x gap between retailer prep and shopper willingness.
The review step matters, too. Giving shoppers a final say raises willingness by half, from 10% to 15%.
My take: handing an AI your card means trusting it to pick the right item at the right price. One bad order costs real money, and most shoppers have not seen enough proof yet. Prepare for agents, but do not build your holiday plan around them. Keep your product data clean so agents can read it. If you test agent checkout, add a confirmation step before payment.
QUICK UPDATES
Google: Google's Search Analytics API, which pipes Search Console data into your own dashboards, now splits branded from non-branded queries on eligible properties. You can also pull Google Lens and Circle to Search performance, and filter by several countries or devices in one request.
Instagram: DMs are getting a spin-off. Meta is testing Direct, a stand-alone Instagram messaging app, in Turkey and Thailand. Sign-up takes an email or phone number, no Instagram account needed, and Instagram users' chats sync across both apps.
LinkedIn: Soon your LinkedIn analytics will talk money. A new monetization tab will join the content and audience sections, listing the platform's earning options and how to sign up. It will also show daily prospect suggestions, in-stream ad opportunities and how your profile performs in LinkedIn's Creator Marketplace.
Snapchat: Luxury sales may close in chats you will never see. In one survey, 54% of luxury shoppers say private conversations with close connections sway their choices more than public feeds. 62% say private chats directly drive purchase decisions. So give them something worth forwarding.
Shopify: Sessions down since November 16? Some of your shoppers were bots. Shopify Analytics now spots more bot traffic in cart and checkout activity, so reports with the human filter may show a small session dip. The change leaves orders, sales and customer data alone. Do not mistake the dip for a BFCM slowdown.