On August 26, 2026, Search Engine Roundtable’s Barry Schwartz reported a new “Product titles” report surfacing inside Google Ads, spotted by Yash Mandlesha and shared on LinkedIn. The report lists product titles that Google AI wrote and served in place of the titles advertisers submitted — and for Michigan’s fitment-driven catalogs, a rewritten title is a merchandising problem long before it is a reporting one.
What is the Google Ads product titles report?
It is a beta report that shows a sample of the product titles Google AI generated and served in your Shopping ads, next to the original titles you supplied. Per the screenshot Mandlesha published, it carries seven columns: the AI-created title, your original title, impressions, product clicks, CTR, cost, and average CPC.
Google’s in-product copy is direct about the trade: “This report shows a sample of product titles created by Google AI that showed in your ads. The product titles you provided will still be shown in ads when they’re the most relevant.” Titles only populate the report once they have accumulated enough impressions, so a thin account may see nothing for weeks while a large catalog fills the table quickly.
Two caveats. This is a sighting in live accounts, not an announced launch — treat the rollout as unconfirmed and check your own account. And it is a different report from the product title A/B experiments feature that shipped earlier in 2026. Experiments test your title against your other title. This one reports on a title you never wrote.
How does Google decide to replace your product title?
The generation engine is text customization — the setting formerly called Automatically Created Assets — which builds ad copy from your landing page, feed attributes, existing ad text, and the query context of the auction. At serve time the system predicts which version is more relevant to the specific query and shows that one.
The mechanism matters because it tells you exactly where substitutions will cluster. The prediction target is query relevance, not catalog correctness. On a narrow query — a part number, a year-make-model-engine string — your original title already matches the query token for token, so the model has nothing to gain and your title serves. On a broad query like “performance clutch kit,” a 140-character fitment title is mostly noise relative to the query, and a shorter AI title scores better.
Run that forward to a case nobody is writing about: a marine dealer in Holland selling raw-water impellers should expect substitutions to concentrate almost entirely on generic “boat water pump impeller” searches and to be nearly absent on “Sierra 18-3082.” If your report shows the opposite pattern, something in your feed is wrong — your part numbers are probably not in the title string where the matching engine can see them.
Why does this hit Michigan catalogs harder than most?
Because a disproportionate share of this state’s e-commerce revenue is fitment-gated. Automotive aftermarket sellers in Warren and Sterling Heights, marine suppliers along the lakeshore, powersports and snowmobile dealers up north, and industrial MRO distributors in Livonia and Grand Rapids all sell catalogs where the buyer is not choosing a product — they are matching one.
In a fitment catalog, the title is not marketing copy. It is the compatibility contract. Merchant Center allows 150 characters and the working formula is brand, then product type, then key attributes, then fitment: “K&N Performance Air Filter fits 2015-2023 Ford F-150 3.5L EcoBoost.” Every one of those tokens is doing load-bearing work, and the ones most likely to get compressed out of an AI-shortened title — the year range, the engine displacement, the submodel — are the exact tokens that decide whether the part bolts on.
A general-merchandise retailer whose title gets tightened from 132 characters to 78 loses some keyword surface. A driveline supplier whose title loses “5.0L” sells a clutch kit to somebody with a 2.7L.
Does a higher CTR on an AI title actually mean more profit?
Not reliably, and the report’s own column list is the tell. It ends at average CPC. There is no returns column, no refund rate, no contribution margin, no RMA count — because Google Ads cannot see any of those.
For a commodity where any unit satisfies the buyer, CTR is a fair proxy for value and the substitution logic is probably doing you a favor. For fitment goods it measures the half of the transaction Google can observe while the expensive half happens three weeks later in your warehouse. Do the arithmetic on a $260 clutch kit at 30% margin: roughly $78 of contribution per sale. One wrong-fitment return costs you that $78 plus freight on a 30-pound assembly plus the labor to inspect and restock — call it $110 all in. If a de-specified title lifts CTR by 0.4 points but adds three returns per hundred incremental clicks, you have paid $330 to buy traffic that looked better in the report.
This is the same measurement gap we flagged when Merchant Center reporting changed on August 24: a number moved without the business moving. Here it is the reverse, and more dangerous — the number improves while the business quietly gets worse.
How should a small team audit AI product titles this week?
All seven steps are doable by a three-person team inside Google Ads, Merchant Center, and one spreadsheet. Budget about four hours.
- Check whether the Product titles report exists in your account. Look under the Shopping or campaign reporting menus. If it is not there yet, the rest of this list still applies — set a calendar reminder to re-check weekly rather than assuming you are exempt.
- Export every row and add three columns of your own: fitment tokens in original, fitment tokens in AI title, and tokens lost. Count year ranges, engine sizes, submodels, and OEM or interchange part numbers. Anything with a non-zero “tokens lost” value is your triage queue.
- Sort that queue by cost, not by CTR delta. The lift is a percentage; the exposure is dollars. A 0.2-point CTR gain on your highest-spend SKU outranks a 2-point gain on a slow mover.
- Open campaign settings and write text guidelines. This is the actual control surface, available across AI Max for Search and Performance Max campaigns. You get up to 25 term exclusions per campaign and up to 40 messaging restrictions at 300 characters each.
- Use messaging restrictions to state the rule, not the wish. Write something like “Never remove model year ranges, engine displacement, submodel, or part numbers from product titles. Fitment data must be preserved verbatim.” Use term exclusions for the vague qualifiers that creep into generated copy — “universal,” “fits most,” “compatible” without a named vehicle.
- Fix the feed so the AI has less reason to intervene. Front-load brand and product type, put fitment in the back half, and use the
structured_titleattribute for any title your own systems generate with AI. If you maintain ACES and PIES data, make sure the fitment strings that reach your title field come from it rather than from marketing copy. - Instrument the outcome Google cannot see. Tag the affected SKUs, then pull return rate for those SKUs on a 30-day trailing window and compare it to the 30 days before substitutions appeared. That single comparison is worth more than the entire CTR column.
What dates matter between now and the holiday ramp?
- Now through mid-September — the audit window. Return data lags sales by weeks, so a change made in October cannot be measured before Black Friday.
- September 4, 2026 — Performance Max AI video reframing begins reshaping creative. Same pattern as titles: Google edits the asset, you find out from a report.
- September 30, 2026 — Google’s updated alcohol advertising policy takes effect, per Search Engine Roundtable’s August 28 report. Relevant to Michigan brewery and distillery e-commerce, not to parts sellers.
- October 1, 2026 — Local Services Ads begins charging for qualifying missed and follow-up calls, relevant to the service side of multi-line businesses.
- November 27 and November 30, 2026 — Black Friday and Cyber Monday. Freeze feed structure by mid-November; do not let a title experiment run during peak.
Frequently asked questions
Can I turn off AI-generated product titles entirely?
There is no single documented kill switch for generated titles specifically. The practical controls are the text customization setting that powers the generation and the campaign-level text guidelines that constrain its output. Term exclusions and messaging restrictions give you a hard “never say this” list, which for fitment catalogs is closer to what you actually need than an on/off toggle would be.
Is the product titles report the same as product title experiments?
No. Title experiments, which Search Engine Roundtable covered earlier in 2026, A/B test two titles you control. The product titles report is a disclosure surface: it tells you after the fact which titles Google AI wrote and served on your behalf. One is a test you designed; the other is a change you are being informed about.
Why is my report empty?
Google states that AI-created titles appear only once they have received enough impressions. A low-volume account, a narrow date range, or a catalog where your original titles keep winning the relevance comparison will all produce an empty table. Widen the date range before concluding nothing has been rewritten.
Do text guidelines apply retroactively?
Not to historical reporting. Google’s own note warns that if you changed guidelines inside the selected date period, you may still see ad copy in the report that violates the current rules even though it would not serve today. Read the report as a record of what ran, then verify separately that new guidelines are shaping what runs now.
The short version
Google shipped visibility, not control, and it arrived in a table whose last column is average CPC. If you sell anything where the wrong part means a return rather than a shrug, export the report this week and write your text guidelines before September. Then judge the outcome on return rate — the number the report will never show you.
Comments are closed