Weekly SEO and Marketing Recap: Sep 11, 2026

Hey — Ali here 👋

Two conferences down, one big model drop, and a week where Google gave with one hand and clawed back with the other. Here's what actually moved this week — and what to do about it before Monday.

LET’S GET STARTED →

Lead Story

GPT-6 Astra Just Landed — And ChatGPT Ads Got a Buy Path You Can Actually Use

OpenAI's newest model is here. GPT-6 Astra launched on 3 Sep, with Work, Codex, and API access following on 9 Sep. It replaces GPT-5.6 Sol, runs across every ChatGPT tier plus Azure and AWS Bedrock, and comes in at $10/$50 per million tokens (input/output). Enterprise access is off by default — you have to flip it on.

Why it matters for operators: Astra is measurably stronger at computer use and browsing — think SERP audits, feed QA, creative brief variants, and site walkthroughs — but OpenAI hasn't published any ChatGPT Search citation changes tied to the new model. Don't invent a ranking story. The shopping model can shift behind the scenes, so watch your diffs.

Now the money thread: ChatGPT Ads.

Self-serve is officially available in India, plus 52 countries total — US, UK, most of the EU, MENA, Brazil, Mexico, Australia, Japan, Korea, and more. Your legal entity has to sit in a listed country. Minimum daily budgets: ₹725 / $25 / €15 / £15. That daily number is a 7-day average — any single day can spike to 2× — and you can bid on CPM, CPC, or oCPC.

The Ads Manager now has a plugin inside ChatGPT and Codex: feed it a brief or site, get ad variants, troubleshoot, preview, and confirm before anything goes live. Audience tools got an upgrade too — you can edit custom audiences without recreating them, set exclusions below the old 25k floor, and pull in GAID and more hashed CAPI fields.

Third buy path: Amazon DSP into ChatGPT. A US pilot launched 10 Sep lets select Amazon advertisers extend campaigns into ChatGPT through Amazon's DSP. Amazon handles setup; OpenAI controls delivery. The units are text or image ads beneath responses, labeled as sponsored, and they support product-feed creatives straight from your catalog.

The measurement gap is still real. Six months into ChatGPT Ads, there's still no published benchmark for what a "good" CPC looks like. Reported numbers swing from under $3 to $10–$13, with outliers near $23. OpenAI's suggested $3–$5 starting max CPC is a floor hint, not an average. Separately, OpenAI is reportedly blocking competitor image- and audio-generation AI tools from advertising on the platform — video-gen ads are still allowed for now.

What to do: Turn on Astra in Work/Codex for research and creative variants — always confirm before the plugin writes to Ads Manager. If you're in India or any Available country, verify your entity and test above the daily floor with your own CAC kill rule — ₹725 buys you almost nothing in learning. Amazon DSP operators in the US have the lowest-friction path into ChatGPT inventory right now. And if you're in commerce, feed completeness matters more than another blog post — roughly 65% of ChatGPT Shopping product recommendations are pulling from structured feeds.

👉 OpenAI GPT-6 Astra → · Astra for Work → · Ads Availability → · SEL Ads stack → · SEL Amazon DSP →

🔍 SEO

Exact and Phrase Match Just Got a Door Into AI Mode

Google confirmed a small experiment: traditional Search campaigns using exact and phrase match can now serve text ads in AI Mode — but only when there's explicit, direct user intent. You don't need to flip every money campaign into AI Max or PMax to show up there. Messy conversational queries still route through the usual channels. If you're running exact or phrase on your best terms, screenshot any new AI Mode impressions now — before you broaden match types to "feed" a test you didn't ask for.

👉 Search Engine Land →

People Also Ask Is Almost Fully AI Overviews Now

A sample of nearly 19.2 million English queries found that 97% of People Also Ask answers in early September were AI Overviews — up from 86% in August and just 12% fourteen months ago. PAA is no longer a traditional snippet surface. If you're still treating it as an FAQ schema play, you're optimizing for a format that barely exists anymore. Citation quality and entity work are what get you into AI Overviews.

👉 SERoundtable →

“Position 1–10” Is Hard to Map — Content “Strong Results” Hit a 12-Year Low

John Mueller on generative AI impressions in Search Console: there are so many interaction paths now that mapping the old position 1–10 framework is increasingly hard to make useful. The Gen-AI report doesn't break out position separately — so if you're still pricing AI visibility using classic average-position logic, you're working with a ruler that doesn't fit the wall.

👉 SER position →

💰 Media Buying

Google Ads Credits Invalidated After Advertisers Spent Them

This one stings. Multiple accounts received what looked like a generous Google Ads credit, spent against it — one case hit $3,200 — and then the credit showed as Invalidated more than a month later. No warning, no clear appeal path. Google's liaison has passed it to the team, but there's no resolution yet. The takeaway is simple: screenshot the offer terms and your billing summary the day you accept any promo credit. Treat it as provisional money until it survives the clawback window.

👉 SERoundtable →

Limited Ad Serving Still Quietly Cuts Affiliates Off Auctions

Limited Ad Serving isn't a disapproval — it's worse in some ways because it's invisible. Google can silently shrink your auction eligibility without a single policy flag. One case study: an authorized retailer spending $3M saw monthly impressions drop from roughly 800k to 350k — a 56% cut — and the appeal process took months. Highest risk: affiliates, insurance, lead-gen, travel, and franchises bidding on brand keywords. Complete your Advertiser Verification, keep disapprovals at zero, make your brand unmistakable on both ads and landing pages, and if you do appeal, include contracts and a walkthrough video.

👉 Search Engine Land →

Snapchat Puts Dynamic Product Ads Inside Sponsored Snaps

Snap is expanding Dynamic Product Ads into Sponsored Snaps — the chat-based format that reaches 85% of active Snapchatters. Early claims include a 22% conversion lift over other placements, and they're testing a Value Optimization objective in alpha. Could be a useful Q4 test if your catalog feed is clean, but chat-based ad fatigue is real — relevance has to be tight or you'll burn the format fast.

👉 Social Media Today →

Publishers Are Buying the Search Traffic They Used to Earn

The top 100 media publishers spent an estimated $113 million on paid search in July 2026 — up 41% year-over-year and 274% over three years. Forbes alone accounted for roughly $72 million in a single month. Commerce keywords are clearing $50 CPCs. The organic moat is gone for big publishers, and they know it. Your edge as a smaller operator isn't history or authority — it's conversion path efficiency and margin discipline.

Also worth noting: Google rolled out MCC manager account dashboards, AI Dashboards you can build via text prompts, and a Search experiments beta for testing budget, tCPA, and tROAS changes. AI Overviews may be squeezing Shopping ad impressions. YouTube alcohol ad personalization opens 30 Oct (still restricted in Egypt, India, Indonesia, and Poland).

👉 Adweek → · SEL AIO vs Shopping →

🤖 AI

Meta Ships Muse, a Personal AI Agent

Meta launched Muse, a personal AI agent, in the US across iOS, Android, and muse.ai. It runs tasks inside a secure VM, works through a messaging-style interface (including WhatsApp), and handles checkout via Link by Stripe. Meta says Muse conversation and VM data are not fed into their ad targeting systems. Free and paid tiers are live.

The affiliate angle: agent-initiated traffic won't always look like a normal user session in your tracker. Attribution and coupon leakage are the immediate risks — if Muse is browsing, comparing, and checking out on behalf of a user, your UTMs and click IDs may not fire the way you expect. Worth watching closely before Q4.

👉 Meta Newsroom → · SMT →

Merchant Center AI Adds Intent/Terms/Attributes — ChatGPT Shopping Leans on Feeds

Google's Merchant Center AI Performance Insights now includes AI Search intent, AI Search terms, and AI attributes — use these alongside Search Console's generative AI tab to see how your products surface in AI-driven results.

On the ChatGPT side, Shopping recommendations lean heavily on product feeds. About 65% of tracked product recommendations are pulling from feed-integrated sources. OpenAI says these are from structured data and are "not ads" — Shopify Catalog and Etsy are connected, other merchants are waitlisted. Feed completeness is the lever. And don't confuse the sponsored ChatGPT ad units with the organic product cards — they're different systems.

👉 SERoundtable → · SEJ Shopping feeds →

📅 Event Spotlight — TES Affiliate Conferences Prague

TES Prague — Diplomat Energy, Two Weeks Out

When: 25–28 Sep 2026
Where: Diplomat Hotel Prague, Evropská 370/15, Prague, Czech Republic
Crowd: ~1,700–2,000 — family-style TES, twice a year (next Spain edition: Seville 8–11 Mar 2027)

Post-Cancún hall for media buyers, networks, and operators who close partnerships. Two weeks to lock flights — not two months.

Code AFFNINJA10 = 10% off on the AffNinja conferences table only — not a TES homepage banner. I will not invent a code that is not on our table.

Let’s grab a coffee in Prague — find me / the AffNinja crew at the Diplomat bar (or a Pilsner). No fancy RSVP link. Buy the ticket, apply the code, come find us.

👉 Get TES tickets → · AffNinja conferences (AFFNINJA10) → · TES guide →

Also This Month

Agentic AI & Automation · 15 Sep · Stockholm · 👉 Info →

AWSummit Bucharest · 21–23 Sep · Bucharest · 👉 Info →

ANS Lisbon 2026 · 29 Sep · Lisbon · 👉 Info →

🧠 Quick Quiz

Answers sit at the bottom — no scrolling ahead

Q1: OpenAI's new flagship launched 3 Sep — but the name has been getting mangled everywhere. Which one is actually correct?
A) GPT Astra 6
B) GPT-6 Astra
C) GPT-5 Astra

Q2: ChatGPT Ads self-serve went live in India this month. What's the minimum daily budget in INR — and is it even worth testing at that floor?
A) Waitlist only · ₹100/day
B) Available · ₹725/day
C) Agency-only forever · no floor published

😂 Meme of the Week

When the platform says 'ads are live' but nobody knows what a good CPC looks like

⚡ Quick Win of the Week

1. Enable GPT-6 Astra on Work/Codex for one research pass + creative variants — human confirm on Ads Manager writes.

2. India / Available-country buyers: confirm entity country, test above ₹725 / $25 with your own CAC kill rule.

3. Exact/phrase Search: screenshot any new AI Mode exposure before changing match types.

4. Ads credits: archive offer + billing screenshot the day you accept.

5. Brand-term affiliates: check Limited Ad Serving with support — hiding the banner does not clear it.

6. Shopping / ChatGPT commerce: Merchant Center AI intent/terms/attributes next to GSC AI impressions; audit feed completeness.

7. TES: apply AFFNINJA10 this week while flights are still sane.

📰 From the AffNinja Blog

AdPlexity Coupon Deep Dive — Native $249 → $169 for Life

If you're running native ads on Mgid, Taboola, or similar networks, AdPlexity Native is the spy tool worth paying for — and right now you don't have to pay full price. Code BLOGECLIPSE_N drops it from $249/mo to $169/mo for life. That's roughly $960 saved over a year, and it locks in permanently.

Running other channels? The deep dive covers every AdPlexity product: Mobile, Desktop, and Adult at $149/mo, Push at $129/mo, YouTube at 30% off, and Social for $49 your first month. One code per checkout — pair with annual billing for the best price, but don't try stacking two codes.

If you only buy one, buy Native.

👉 AdPlexity coupons breakdown →