Advertising in Google AI Mode: The 2026 Guide

Over four weeks in late summer 2026, Google quietly widened the set of campaigns whose ads can appear inside AI Mode, the Gemini-powered conversational surface in Search. Shopping products came first, then exact and phrase match keywords from ordinary Search campaigns. Advertising in Google AI Mode is no longer something reserved for AI Max and Performance Max: most accounts running automated bidding are already eligible, with no separate line in reporting and no documented way to opt out. This guide covers what actually changed, how to tell whether AI Mode traffic is already in your account, which levers you still hold, and why your product data — not your bid — now decides whether Gemini builds your product into an answer.

What AI Mode is and why it stopped being an experiment

AI Mode replaces the ten blue links with a generated answer: reasoning, comparisons, and recommendation blocks. The user refines by conversation — “which of these works in a small apartment,” “anything quieter,” “what can arrive by Monday.” Every refinement is a new query, and none of them exist in your keyword planner.

Scale is no longer theoretical. Google put AI Mode past a billion monthly users by I/O 2026, with roughly 30% of ad-eligible queries already showing ads. For a typical account that means AI Mode is not a future channel — it is a slice of current traffic that the interface simply does not label.

The core shift of 2026: an ad in AI Mode is not served, it is assembled. Gemini takes your assets, your product feed and your landing page and builds a response around the user’s exact phrasing. You supply raw material, not finished advertising.

Advertising in Google AI Mode: which campaigns are eligible in 2026

There is no direct targeting. You cannot create an “AI Mode campaign” or select it as a placement. Eligibility follows from campaign type and settings.

Traffic source	Eligible for AI Mode	Condition
Performance Max	Yes	Automated bidding, feed or asset groups
AI Max for Search	Yes	Full access, including the newer formats
Shopping campaigns	Yes	Data pulled from Merchant Center
Broad match with Smart Bidding	Yes	Baseline condition since early 2026
Exact and phrase match	Yes, since September 2026	Only queries with explicit, direct intent
Manual CPC	No	Smart Bidding is a requirement, not a suggestion

That last row deserves attention. If you still keep part of the account on manual bidding for control reasons, that choice now carries an explicit cost: exclusion from a growing surface. What does and does not disturb an automated strategy is covered in our piece on the Smart Bidding learning period.

August and September 2026: the timeline that matters

Date	Change	Practical meaning
August 31, 2026	Shopping products inside AI Mode	A product carousel appears in the body of the answer; Gemini selects items from your Merchant Center feed
September 1, 2026	AI Max auto-migration goes live	Search campaigns using automatically created assets or broad match were upgraded, with text customization on by default
September 3–4, 2026	Exact and phrase match in AI Mode	Standard Search campaigns became eligible on direct-intent queries
September 16, 2026	Conversational attributes in Merchant Center	Feeds can carry question-and-answer pairs and related-product data for dialogue scenarios

Google’s own wording on the September expansion was deliberately narrow: exact and phrase match keywords are eligible to serve text ads in AI Mode “limited to cases where there is explicit and direct user intent.” No geography, no vertical restriction, no account share and no end date were published, and no opt-out was documented.

The four formats from Marketing Live 2026

Conversational Discovery Ads — surface as the shopper describes a need; Gemini writes the bridge between the question and your product from what it knows about that product.
Highlighted Answers — your product sits inside the recommendation set beside organic suggestions. Available only to AI Max and Performance Max.
AI-powered Shopping Ads — these run on Search rather than in AI Mode, with Gemini writing explainers for considered purchases such as appliances.
Business Agent — a chat agent replacing lead forms, expanded to YouTube in a US beta in September 2026.

The practical takeaway: a standard Search campaign will not reach Highlighted Answers no matter how good the ad copy is. That surface requires AI Max or PMax. What the migration changes in day-to-day control is laid out in AI Max for Search in 2026.

Why data quality beats bidding in this surface

In a classic auction you wrote the ad in advance and Google chose between finished texts. In AI Mode there is no finished text. The model composes the response on the fly from three sources: the product feed, campaign assets, and the landing page.

The consequence is direct — completeness and accuracy of product data became a ranking input. A product with material, sizing, compatibility, expected lifespan and answers to common pre-purchase questions gives the model something to justify a recommendation with. A product with a title, a price and a photo gives it nothing, and loses to a richer competitor at the same bid or a higher one.

For teams this moves effort. The lever used to be bids and campaign structure; a large share of the outcome now happens in Merchant Center. The mechanics of rewriting titles and shipping changes reversibly are in feed rules and supplemental feeds.

Conversational attributes: what to fill in first

Question-and-answer pairs. The heaviest single input. Source them from real support tickets and pre-sale chats, not from imagination. Five to eight pairs on your top SKUs is already meaningful coverage.
Related products. What pairs with what, what replaces what, what needs buying alongside. This feeds the “what else do I need” branch of a conversation.
Variants and item groups. A correct item_group_id stops the model from presenting one product five times as five options.
Material, fit, durability, compatibility. Everything a buyer asks before committing to an expensive item.
Popularity signals. Which SKUs actually sell, so the model does not surface dead stock.

Ship all of this through a supplemental feed. Live listings stay untouched, and an experiment can be rolled back in a minute.

The reporting problem nobody can solve for you

Say this plainly, because half the wrong conclusions this autumn start here: Google Ads has no segment, column or filter that separates AI Mode delivery from regular Search delivery. An exact match keyword row can now aggregate literal matches, close variants and AI Mode impressions into a single number covering three different user behaviours.

A CTR drop in September does not prove your ads got worse. A new surface may simply have joined the same row.
Month-over-month comparisons crossing September 1 compare two different traffic mixes, not two levels of performance. Annotate the date.
Deriving max bids from CTR is unreliable in this window. Work from conversions and value instead.

Performance Max created the same kind of blind spot, and the answer there was never a single report — it was a set of proxy cuts. The approach transfers; see Performance Max channel reporting.

Proxy metrics that actually tell you something

Signal	Where to find it	What it indicates
AI Max conversational query report	AI Max campaigns	Closest available view of dialogue-shaped queries
Share of long, unusual search terms	Search terms report	Indirect marker of conversational traffic
Branded query trend	Search terms plus Search Console	Lagging indicator — AI Mode often creates awareness without a click
Impression-to-click gap	Custom columns	Appearing in an answer does not always end in a site visit
Assisted conversions and lag	GA4, conversion lag reports	Conversational journeys run longer than classic ones

Link Search Console while you are at it: part of the AI Mode effect shows up in organic impressions and branded demand before it shows up in paid metrics. How to read that view is covered in today’s companion piece on the paid and organic report.

Seven things to do with your Search campaigns now

Freeze a pre-September baseline. Export CTR, CPC, CVR and impression share for August. Without it you cannot separate AI Mode from seasonality.
Annotate August 31 and September 1 in every dashboard. In three months nobody will remember why the lines break there.
Identify which campaigns were auto-migrated. Change history shows it. Campaigns with automatically created assets or broad match are the first candidates.
Rebuild account-level negative keywords. Conversational phrasing produces queries that were never in your keyword set. B2B and service accounts feel this fastest.
Fence off branded traffic explicitly. Brand lists and exclusions work here, but they have to be configured — the default behaviour happily absorbs brand demand. See Google Ads brand lists.
Constrain generated copy. If text customization switched on automatically, set term exclusions and messaging restrictions. Details in text guidelines for AI-written ad copy.
Start on the feed. Fifty revenue-leading SKUs first: titles, descriptions, Q&A pairs, variant attributes.

A 30-day plan

Days 1–3. Audit: which campaigns migrated, where text customization is active, which strategies are still manual. Export the August baseline.

Days 4–10. Feed work on the top 50 SKUs: natural-language titles, complete attributes, five to eight Q&A pairs drawn from support history. All via supplemental feed.

Days 11–17. Defence: account-level negatives, brand lists, text guidelines, and a landing-page check against what the assets promise.

Days 18–24. Reporting: custom columns for long-query share, impression-to-click gap, assisted conversions. Link Search Console if it is not linked already.

Days 25–30. First read: compare to baseline, review new query shapes, decide whether to extend feed work to the next 200 SKUs.

What AI Mode does not change

Your economics. Maximum allowable CPA still comes from margin and LTV, not from how elegantly Gemini described the product. If that number is fuzzy, start with media buying unit economics.
Landing page obligations. The page must load, must be crawlable, and must deliver what the answer promised.
Conversion signal quality. The model learns from what you send it. Junk conversions produce junk decisions in this surface exactly as in every other.
The need for a healthy ad account. None of the new formats are available to an account under serving restrictions.

Five persistent misconceptions

“You can switch AI Mode off.” No documented opt-out exists for text ads in AI Mode. You control campaign type and bidding strategy; that is not the same as a toggle.

“This is an ecommerce story.” The product formats mostly are. Text ads on exact and phrase match are all of Search — services, B2B and lead generation included.

“Move to AI Max and you are done.” AI Max unlocks formats; the model still assembles answers from your data. An empty product record stays empty in any campaign type.

“No report, nothing to measure.” Conversions, value and incremental effect are all still measurable. Only the AI-Mode-versus-Search cut is missing, which is an argument for stronger outcome measurement, not for giving up.

“Rebuild everything immediately.” No. Freeze a baseline, protect brand and budget, then improve product data methodically. Aggressive September edits only blend the migration effect with your own.

Ten-point checklist

August baseline exported and stored separately.
August 31 and September 1 annotated in every dashboard.
List of auto-migrated campaigns documented.
Text customization status known; text guidelines configured.
Account-level negatives refreshed for conversational phrasing.
Brand lists and exclusions explicitly set.
Top 50 SKUs enriched in a supplemental feed, Q&A pairs included.
Variant attributes and item_group_id verified — no duplicate presentation.
Search Console linked; branded demand tracked.
Any remaining manual-bidding campaigns kept manual on purpose, not by inertia.

Strip away the Gemini branding and the shift is simple: ad systems increasingly paraphrase your data instead of displaying your copy. The advertiser with more accurate, more complete data wins — not the one bidding loudest.

Where this meets account infrastructure

Access to the new formats depends on a healthy account with automated bidding and history. An account under limited serving, with an unverified advertiser or a wiped history, simply does not enter these auctions. If infrastructure is the bottleneck, look at Google Ads agency accounts and advertiser verification. For a review of your own campaign structure against the new surfaces, see the full list of PPC Rebels services.

FAQ

Can I opt out of AI Mode?

No documented switch exists. You influence eligibility indirectly through campaign type and bidding strategy, since manual CPC is excluded. Reverting to manual bidding purely to avoid AI Mode is a poor trade — you lose more in optimisation quality than you gain in control.

How much of my traffic comes from AI Mode?

There is no exact answer available in the interface. Approximate it with the AI Max conversational query report, the share of long unusual search terms, the impression-to-click gap, and branded demand trends.

Does exact match now behave like broad match?

No. Matching logic is unchanged; the serving surface expanded. Exact and phrase keywords became eligible in AI Mode on direct-intent queries, which makes the reporting row less homogeneous without turning it into broad match.

Do I need Merchant Center if I sell services?

For product formats, yes. For text ads in AI Mode, no — the model draws on campaign assets and the landing page. Which means for services the quality of assets and the factual depth of your site carry the weight.

Should every campaign move to AI Max?

Not automatically. AI Max unlocks Highlighted Answers and text customization but takes back some control. Test it on a subset through an experiment and judge on conversions and value, not CTR.

What does this do to budgets?

AI Mode has no separate budget; it is an additional surface inside existing campaigns. In practice some accounts see more impressions with a longer path to conversion, which puts pressure on daily budgets before returns show up.

My CTR dropped after September 1. What now?

Check the traffic mix before rewriting ads. If impressions rose sharply while clicks rose less, a new surface most likely joined the same row and the CTR change is a composition effect, not a copy problem.

How quickly does conversational attribute work pay off?

Two to four weeks for visible movement on top SKUs is a reasonable expectation, provided the attributes are substantive rather than filled in for completeness. Treat that as a benchmark, not a promise — category impression volume drives the timeline.

Do I need separate creative for AI Mode?

There is no separate upload for this surface. But the bar for assets rises: specific, factual copy gives the model something to build with, and generic phrasing performs worse than it used to.

Feed work or query defence first?

Defence. Negatives and brand rules stop losses immediately, while feed enrichment pays off later. One week on defence, then a month on data, is a workable order.

Does AI Mode affect organic traffic?

Yes — a generated answer can close the question without a site visit. That is precisely why paid and organic impressions are worth reading together in 2026 rather than in separate meetings.

Where should a one-person team start?

Three things in week one: annotations in reporting, account-level negatives, and enrichment of the twenty highest-revenue SKUs. Everything else can wait without much risk.

Advertising in Google AI Mode in 2026: How Your Ads Reach Gemini Answers