Dynamic Creative Optimization in Google Ads 2026: DCO setup and strategy

For most of Google Ads' history, an ad was a fixed object: you wrote a headline, paired it with an image, and that exact unit served to everyone. Dynamic Creative Optimization inverts this. Instead of building finished ads, you supply raw materials — headlines, descriptions, images, logos — plus a data feed, and Google's machine learning assembles the combination most likely to convert for each user and context. The ad a shopper sees is composed in real time, often showing the precise product they browsed at its current price.

This tutorial is for marketers and Google Ads practitioners who want to implement DCO properly rather than just clicking 'create responsive ad' and hoping. We will cover what DCO actually encompasses in 2026, the building blocks (feeds, assets, the combination engine), step-by-step dynamic remarketing setup, responsive display assets, ad customizers for Search, feed construction, testing creative at scale, and the measurement approach that makes sense when a machine, not you, decides which combination serves. The recurring theme: your job moves from crafting ads to curating inputs.

What DCO actually means in Google Ads in 2026

DCO is often discussed as if it were a single feature, but in Google Ads it is a set of overlapping capabilities that share a common principle: automated assembly and optimization of creative from component parts. Understanding which capability does what is the first step to using them deliberately.

The three pillars of DCO in Google Ads:

• Responsive ads (display and search). You supply multiple headlines, descriptions, images, and logos; Google mixes them into combinations and learns which perform best for each context. This is DCO applied to your own creative assets, no product feed required.
• Dynamic remarketing. Tied to a data feed, this shows users the exact items they viewed — with live price, image, and availability — wrapped in responsive creative. This is the most powerful and personalized form of DCO, and it requires a feed.
• Ad customizers (Search). These inject dynamic values — price, date, location, countdown, inventory count — into Search ad text at serve time, from a data source you upload, so one ad template renders many context-specific variants.

Most mature accounts use all three in concert: responsive assets carry prospecting and brand campaigns, dynamic remarketing personalizes re-engagement, and customizers keep high-SKU or time-sensitive Search ads always current. They are complementary, not competing.

What unites them is the move from a one-ad model to an ingredients-plus-rules model. You provide the ingredients (assets and feed data) and the goal (a conversion action), and Google's machine learning composes and tests creative against that goal continuously, at a granularity — per impression, per user, per context — that manual workflows cannot approach. The payoff is personalization and testing velocity; the cost is control, which you manage through asset curation, light pinning, and feed governance rather than through hand-built ads. For the broader context of how automated creative fits alongside Google's automated campaign types, our complete Performance Max guide is a useful companion, since Performance Max leans heavily on the same asset-combination engine.

The building blocks: feeds, assets, and the combination engine

Every DCO implementation rests on three components, and the quality of each one caps the performance of the whole. Weakness in any single building block bottlenecks the others, no matter how good they are.

The asset pool. This is your library of creative components: headlines, descriptions, images, logos, and videos. The engine can only test combinations of what you give it, so the pool's variety — not just its size — determines the ceiling. Twenty near-identical headlines give the machine nothing to learn from; five genuinely different angles (benefit, offer, feature, social proof, urgency) give it real material. Asset quality and diversity are the inputs you most directly control.

The data feed. For feed-driven DCO (dynamic remarketing especially), the feed supplies the live content: products, their prices, images, availability, and URLs. Feed quality is decisive because dynamic ads inherit the feed's accuracy — a stale price or a broken image URL becomes a bad ad shown to a high-intent user. The feed must be complete (all required attributes), current (refreshed at least daily), and clean (valid images, accurate prices, no disapprovals).

The combination engine. This is Google's machine learning that assembles assets and feed data into the optimal ad per impression and optimizes against your conversion goal. You do not control it directly; you influence it through the inputs (assets, feed), the constraints (pinning), and the objective (conversion action and bidding strategy). It needs conversion volume and time to learn, which is why DCO underperforms on tiny accounts with sparse conversion data.

The practical implication is a division of labor: you own the asset pool and the feed, Google owns the combination. Most DCO failures trace back to a neglected input — a thin asset pool with no real variety, or a feed that drifts out of date — rather than to the engine itself. Invest your effort where you have leverage: the ingredients.

Setting up dynamic remarketing step by step

Dynamic remarketing is the highest-impact form of DCO because it shows users the exact items they engaged with, at live prices — and it converts far better than generic remarketing for exactly that reason. Setting it up correctly has four parts that must all line up.

The setup sequence:

• Build the feed. Create the product or custom feed (covered in detail in a later section) with item IDs that exactly match the IDs your site will report when users view items. The match between feed IDs and on-site event IDs is the linchpin — if they do not match, the ad cannot pull the right item.
• Tag the site with dynamic remarketing parameters. Beyond the standard remarketing tag, dynamic remarketing requires custom parameters that report which item the user viewed: the item ID, the page type (home, category, product, cart, purchase), and often a value. These parameters tell Google which feed entry to show that user later.
• Build behavior-segmented audiences. Create remarketing lists segmented by funnel action — viewed an item, added to cart but did not buy, purchased — so you can show different dynamic creative and bids to each. Cart abandoners warrant the most aggressive treatment; recent purchasers may be excluded or shown complementary items.
• Create the dynamic ad. Build a responsive display ad that wraps your assets (headlines, descriptions, logo, layout) around the feed item the engine inserts. Preview it to confirm it correctly displays a viewed product with its live image and price.

The validation step matters more than people expect. Before scaling spend, use the ad preview and the tag diagnostics to confirm the full chain works: a user views item X, the tag reports item X's ID, the audience captures that user, and the dynamic ad later serves item X with its current price and image. A break anywhere in that chain — most often an ID mismatch between feed and tag, or a missing page-type parameter — produces dynamic ads that fall back to generic content, quietly losing the personalization that justifies the whole approach.

A note on goals: dynamic remarketing is usually a high-ROAS, lower-volume channel because it targets warm, already-engaged users. Treat it as a distinct campaign with its own bidding (often Target ROAS) and exclusions, rather than blending it with cold prospecting, so its strong economics are visible and protected. For building the durable audiences that feed remarketing in a cookieless world, see our Customer Match and first-party data guide.

Responsive display assets and automated combinations

Responsive display ads are DCO applied to the Google Display Network: you provide assets, Google assembles and tests combinations across the millions of placements in the network, automatically fitting them to each placement's size and format. This is the most accessible entry point to DCO because it needs no feed — just a well-built asset pool.

What to provide, and how to provide it well:

• Multiple headlines (short and long). Cover different angles — a benefit headline, an offer headline, a feature headline, a social-proof headline — so the engine can match message to context.
• Several descriptions that each stand alone and make sense paired with any headline, since combinations are free-form.
• Images in both landscape (1.91:1) and square (1:1) ratios, with enough variety (product shots, lifestyle, different subjects) to suit different placements.
• Logos in landscape and square, clean and recognizable at small sizes.
• Optional video, which can lift performance on placements that support it.

The cardinal rule is combinatorial coherence: because any headline can pair with any description and image, every asset must work in any combination. Avoid assets that only make sense in a specific sequence ('And that's not all...') or that contradict each other across combinations. This discipline feels constraining to writers used to crafting a single linear ad, but it is what lets the engine recombine freely without producing nonsense.

Pinning deserves careful handling. Google lets you pin assets to fixed positions — useful for a required legal disclaimer or a brand line that must always appear — but over-pinning defeats the purpose by collapsing the combination space the engine optimizes over. The guidance is to pin only what genuinely must be fixed, and to provide at least two or three options even for pinned positions where allowed, so some optimization survives. Treat pinning as a scalpel for compliance and brand requirements, not a default control lever.

Finally, remember that responsive display reaches an enormous, varied inventory, so placement and brand-safety controls matter. Review where ads serve, exclude unsuitable placements and content categories, and use audience signals to steer the engine toward relevant users. The combination engine optimizes creative, but it does not by itself guarantee your ad appears in contexts you would endorse — that remains your responsibility through exclusions and targeting. The same responsive-asset discipline applies on the Search side; for headline-writing depth, our responsive search ads templates guide is a strong reference.

Ad customizers for dynamic Search ads

Ad customizers bring DCO to Search by injecting dynamic values into ad text at serve time. Where responsive search ads optimize the combination of your written assets, customizers change the content of the ad based on context — turning a single ad template into thousands of context-specific variants without manual creation.

How customizers work in practice:

• You upload a data source mapping values to conditions — for example, a price per product, a date per event, a location per region, or an inventory count per item — keyed to keywords, audiences, or schedules.
• You insert customizer fields into your ad text using the customizer syntax, with a default value specified for when no specific match applies.
• At serve time, Google substitutes the right value for the matched condition, so the same ad displays 'From €49' for one keyword and 'From €129' for another, or a live countdown to a sale's end.

The use cases where customizers shine are high-SKU and time-sensitive accounts. An ecommerce retailer with thousands of products can show accurate, current prices in Search ads without building a separate ad per product. An events or travel advertiser can show live countdowns, dates, or 'X left' inventory urgency that updates automatically. A multi-location business can inject the nearest location or region-specific offer. In every case, customizers replace what would otherwise be an unmanageable sprawl of hand-built ad variants with one template plus a data source.

Two cautions keep customizers reliable. First, always set sensible defaults — if the data source lacks a value for a given condition, the default displays, so it must read cleanly as a standalone ad. A missing default can suppress the ad entirely or show an awkward fallback. Second, keep the data source current, exactly as you would a feed: a customizer showing an expired price or a finished countdown is worse than a static ad. Schedule updates and validate that values render correctly across your targeting conditions before scaling.

Customizers and dynamic remarketing share a philosophy — let live data drive creative — applied to different surfaces. Used together with responsive assets, they complete a DCO stack that keeps both Display and Search creative personalized and current. The countdown and live-price mechanics are especially valuable for time-bound verticals; our guides on events and ticketing and travel agencies and OTAs show these dynamic-ad tactics applied in context.

Building and structuring the data feed

The feed is the heart of feed-driven DCO, and its quality directly determines the quality of every dynamic ad it powers. Whether you use a Merchant Center product feed for ecommerce or a custom dynamic remarketing feed for another vertical, the same principles apply: complete, accurate, current, and correctly matched to your site's events.

The required and recommended attributes:

• ID — a unique, stable identifier that must exactly match the item ID your site reports in its dynamic remarketing tag. This match is non-negotiable; mismatched IDs break personalization.
• Title — a clear, accurate item name that renders well in ad creative.
• Price — the current price, kept in sync with your site to avoid bait-and-switch.
• Image URL — a valid, high-quality image that loads reliably; broken images produce broken ads.
• Final URL — the landing page for the item, ideally the exact product or detail page.
• Category and custom attributes — used for grouping, targeting, and showing related items.

For non-ecommerce verticals, Google supports custom dynamic remarketing feeds tailored to the business type — travel (destinations, hotels, flights), real estate (listings), jobs, education, and local deals — each with vertical-appropriate attributes. The structure differs, but the discipline is identical: every item complete, every field accurate, the whole feed refreshed frequently.

Feed governance is an ongoing job, not a one-time setup. Schedule the feed to refresh at least daily — more often for fast-moving prices or availability — and monitor for the three feed killers: disapprovals (items rejected for policy or data issues), coverage gaps (items missing required fields and therefore not serving), and accuracy drift (prices or availability that no longer match the site). A weekly feed-health review catches these before they erode performance. The feed is the input where neglect does the most damage and where attention pays off most directly, because a single feed can power dynamic remarketing, feed-driven Performance Max, and Search customizers simultaneously.

Creative testing at scale without losing control

DCO's promise is testing creative at a scale impossible by hand — but that scale only helps if you feed it the right experiments and read the results correctly. Testing at scale means changing how you think about creative testing, not just doing more of the old way.

The shift from manual A/B to DCO-era testing:

• Test angles, not just variants. Instead of testing two near-identical headlines, give the engine genuinely different messages (benefit vs offer vs social proof) and let it discover which resonates per context. The unit of testing becomes the angle, not the wording tweak.
• Let the engine run the multivariate test. Manually testing every combination of headlines, descriptions, and images is combinatorially impossible; the engine does it continuously. Your job is to ensure the ingredients are diverse enough that the test is meaningful.
• Refresh the pool, do not just prune it. When an asset underperforms, replace it with a new variant rather than only deleting — the engine needs a deep, fresh pool to keep finding winners and to fight fatigue.
• Use experiments for structural questions. Reserve formal Google Ads experiments (campaign-level A/B tests) for bigger questions — DCO vs static, one feed strategy vs another — where you need clean causal read-out, not for individual asset comparisons the engine already handles.

The control you retain is at the level of inputs and constraints. You decide which assets enter the pool, which must be pinned, what the feed contains, and what objective the engine optimizes toward. You do not — and should not try to — micromanage which combination wins; that is the engine's job and it does it faster and more granularly than any human. The skill that matters in the DCO era is curation: assembling a diverse, high-quality asset pool, maintaining a clean feed, setting the right objective, and refreshing inputs on a cadence. Practitioners who cling to controlling individual ads fight the system; those who master input curation get the scale benefit.

A discipline worth keeping is the refresh cadence. Creative fatigue is real even with DCO — audiences tire of an asset pool over weeks. Rotate in new headlines, descriptions, and images regularly (every few weeks for active campaigns), retire consistently Low-rated assets, and keep the feed current. Treat the asset pool as a living garden you tend, not a set-and-forget configuration. This ongoing curation, more than any one-time setup, is what sustains DCO performance over time.

Measuring DCO performance and asset-level signals

Measuring DCO requires a two-level approach, because the machine assembles the ads and you cannot judge performance the old way — by comparing hand-built ads head to head. Instead you measure outcomes at the campaign level and ingredients at the asset level.

Campaign-level: judge by outcomes. The ultimate test of DCO is whether it moves your primary conversion metric — cost per conversion, ROAS, conversion volume — against your goal. Compare DCO campaigns to your benchmarks and, where you want a clean causal read, to a static-creative control via a formal experiment. The combination that won any given impression is not the metric that matters; the aggregate business outcome is. This is liberating once internalized: you stop agonizing over which headline won and start asking whether the campaign hit its target.

Asset-level: read the signals and curate. Google provides asset-performance ratings — typically Low, Good, and Best — plus asset-level reporting that shows which headlines, descriptions, and images are contributing. Use these to guide curation:

• Replace Low-rated assets with new variants to lift the pool's overall quality.
• Study Best-rated assets to understand which angles and visuals resonate, then create more in that vein.
• Maintain pool depth so the engine always has fresh, varied material — do not let pruning thin the pool below what optimization needs.
• Watch for fatigue in assets whose ratings decline over time, and rotate them out on a cadence.

For feed-driven DCO, add feed-health metrics to your measurement: coverage (share of catalog actually serving), disapproval rate, and price/availability accuracy. A feed problem masquerades as a creative problem — dynamic ads underperform because they show stale or missing data, not because the engine failed. Monitoring feed health alongside creative performance separates the two diagnoses.

The overarching measurement principle is the same as the overarching strategic principle: you have moved from optimizing ads to curating inputs, so you measure inputs (asset ratings, feed health) and outcomes (conversions, ROAS), not the intermediate combinations the engine churns through invisibly. Practitioners who try to report on individual combinations drown in noise; those who track the two levels that matter steer DCO effectively.

If you want a second opinion on whether your dynamic ads, feeds, and asset pools are actually pulling their weight — the most common DCO failures are a thin asset pool with no real variety and a feed quietly drifting out of date — SteerAds runs a free 14-day audit on Google and Microsoft Ads that reviews creative, feed health, and asset-level performance end to end.

Sources

• support.google.com/google-ads — Dynamic remarketing setup documentation
• support.google.com/google-ads — Responsive display ads and asset guidance
• support.google.com/google-ads — Ad customizers documentation
• support.google.com/merchants — Merchant Center product feed specification
• thinkwithgoogle.com — Google research on creative optimization and personalization

FAQ