Soku AIBlog
Insights, tutorials, and updates from the Soku AI team.
ChatGPT Ads Identifier Normalization: OpenAI Rewrote the Hashing Rules, and Leading Zeroes Now Matter
OpenAI rewrote the normalization rules for hashed identifiers in the ChatGPT Ads Measurement Pixel and Conversions API. The phone rule changed from removing a leading + or 00 to removing a leading + and any leading zeroes, which changes the hash for every number stored in national format. The exact previous and current wording, a worked example table, the accent rule almost every implementation gets wrong, and what the help centre still does not say. Verified 2026-08-31.
12 min read · August 31, 2026
Five Checks Before You Let an Agent Spend Your Ad Budget
Autonomy is not a slider you turn up. Five specific checks — attribution of cause, hard floors, decision replay, missing-data behaviour, and rollback ownership — that decide whether an agent managing your campaigns compounds your results or your mistakes.
8 min read · August 31, 2026
Your Ad Agent Can't See the Auction It's Bidding Into
A rising CPC caused by creative fatigue looks identical to a rising CPC caused by a new bidder. We measured two real competitive sets to show why an ad agent optimizing inside one account cannot tell them apart.
9 min read · August 28, 2026
ChatGPT Ads Now Lets You Create an Account Before Your Country Opens
OpenAI rewrote the ChatGPT Ads front door on 2026-08-28. The advertiser interest form at openai.com/advertisers is gone from the documentation; you now sign up directly at ads.openai.com from any country. Account creation and self-service availability are formally separate — and OpenAI states plainly that creating an account does not grant early access. What the new access-type table says, which of the 40 listed countries can actually self-serve today, and what a blocked advertiser should do this week. Verified 2026-08-28.
14 min read · August 28, 2026
ChatGPT Ads Custom Audiences Can Now Be Edited — but Only the Developer Docs Know It
On 2026-08-25 OpenAI's Ads API gained Add, Remove, Replace and Merge operations on custom audiences, plus small and empty exclusion-only audiences. The ChatGPT Ads help centre still says custom audiences cannot be edited after creation and still says every audience needs 25,000 matched users. Both pages are live and they contradict each other. What the API actually does, the revision and idempotency rules that keep a membership update from double-applying, and which surface to trust for which job. Verified 2026-08-26.
16 min read · August 26, 2026
ChatGPT Ads Now Rolls Unspent Daily Budget Forward — and OpenAI Finally Fixed Its Own Contradiction
OpenAI updated both its Daily Budgets and Billing help articles on 2026-08-25 with a mechanic neither page carried before: if your campaign underdelivers, the unused budget can be spent later in the same seven-day period. The same edit finally corrected the Billing page, which had been contradicting the Daily Budgets page since late July. What catch-up delivery means for pacing, why an underdelivering campaign is now a deferred liability rather than a saving, and the post-pause reconciliation line almost nobody will read. Verified 2026-08-25.
13 min read · August 25, 2026
ChatGPT Ads Expands to 31 European Countries — But Not Through Ads Manager Yet
OpenAI is bringing ChatGPT Ads to 31 European markets, its largest expansion to date. The help centre now lists all 31 as Coming Soon rather than Available, and the announcement explains why: initial access runs through the OpenAI Ads Solutions team, agency partners and technology partners, with self-service Ads Manager access following later this summer. The full country list, what Coming Soon actually means operationally, and the ad-groups rewrite that shipped the same week. Verified 2026-08-24.
15 min read · August 24, 2026
ChatGPT Ads Adds a Maximize Results Bid Strategy — And It Is Already Selected by Default
OpenAI shipped an automated bid strategy for ChatGPT Ads called Maximize results, and the help centre says it is selected by default for eligible new ad groups. The exact official wording, the cost guarantee OpenAI explicitly declines to make, how it differs from an oCPC Bid Cap, and the split nobody is naming: Maximize results appears in no developer document, so ad groups you create through the Ads API still take a manual bid. Verified 2026-08-21.
15 min read · August 21, 2026
Ad Creative Testing Platforms Compared: What Each One Actually Tests
Most tools sold as ad creative testing platforms do not run the test. They label creative, report on it after the fact, or generate more of it. A category-by-category comparison of what each type actually measures, what it cannot tell you, and how to pick by the decision you need to make. Verified 2026-08-20.
14 min read · August 20, 2026
AI vs Human Campaign Audit Accuracy: A Reproducible Test
Everyone claims AI audits ad accounts as well as a specialist. Almost nobody publishes a protocol you could use to check. Here is the full test design — the seeded-defect method, the scoring rubric, the traps that make naive comparisons meaningless — so you can run it on your own account instead of trusting a vendor's number. Verified 2026-08-20.
16 min read · August 20, 2026
Creative Fatigue Benchmarks: When Meta, TikTok and YouTube Ads Wear Out
There is no universal number of days after which an ad is spent. Fatigue is a function of frequency against audience size, not calendar time — and the three platforms fatigue on different mechanisms. How to compute your own threshold, and the four decay patterns that are not fatigue at all. Verified 2026-08-20.
15 min read · August 20, 2026
AI Ad Management for Small Businesses: A Practical 2026 Guide
AI ad management for small business, without the hype: the budget below which automation cannot learn, the approval rules that keep an agent safe on a small account, what you can actually measure at low conversion volume, and the cases where automation is not worth it. Verified 2026-08-19.
16 min read · August 19, 2026