How to Set Up a Google Demand Gen Campaign Step by Step

Demand Gen needs a conversion action it can optimise toward. In Google Ads, confirm the primary purchase or lead action is recording, imported correctly from GA4 or the tag, and marked as a primary action. Run a test conversion. If the account cannot see conversions, automated bidding is guessing, and your creative test will look like a bid problem you cannot debug.

Check that you are not optimising to a page view or an add-to-cart you accidentally set as primary last quarter. Demand Gen will happily buy cheap add-to-carts if that is the action you crowned, then you will say YouTube 'does not drive revenue.' For ecommerce, purchase with a value is the usual primary; for lead gen, a qualified lead or form submit you actually want more of, not every newsletter tick. Enhanced conversions or server-side tagging help when browser matching is thin — the same reason other platforms want CAPI — so do that work before you argue about hooks.

If conversion volume is very low, you will need more budget or a higher-funnel primary for a while, then switch once purchases exist. Do not start Demand Gen on the same day you install the tag and expect a stable CPA by Friday. Give the account a few real conversions in any campaign type first if you can, even a small search or Performance Max test, so the conversion action is not theoretical. Tracking is the unglamorous half of 'YouTube creative testing,' and skipping it is how good UGC gets paused.

In Google Ads, create a new Demand Gen campaign and choose the conversion goal you just verified. Set location to where you fulfil, language to match the creative, and a daily budget you can hold for at least a week. Do not recycle an old Video campaign structure and hope Shorts inventory shows up as a side effect of a pre-roll setting you never checked.

Demand Gen is the campaign type built around visual asset groups and feed-like placements, including Shorts and YouTube in-feed. Standard Video campaigns are still centred on in-stream and watch-page behaviour, with Shorts treated as an extra rather than the job. If Shorts and Discover are what you want, start in Demand Gen. You can later exclude a placement that wastes spend; you cannot assume a default Video campaign is a Shorts campaign because the UI mentioned YouTube. Read the review screen before you publish.

Name the campaign with market, goal and test number so reporting six weeks later is still human. Set brand exclusions if you do not want to appear on competitor or unsafe inventory, using Google's brand tools rather than a homemade URL list you will forget to maintain. Keep the first campaign simple: one market, one goal, one product or one offer. A Demand Gen campaign that tries to sell the whole catalogue on day one is hard to diagnose, because every asset and every SKU can take the blame for a bad CPA.

An asset group is the creative set the system mixes across placements. Load at least two or three distinct 9:16 videos with different hooks, plus supporting 1:1 or landscape cuts and images if you have them, headlines, descriptions and a logo. One video in one ratio is not an asset group — it is a single ad wearing a new label, and the report will have nothing to teach you.

Use the same testing logic you would on Meta: one locked body script, three hooks, two avatars if you can — six vertical files is a strong start and matches the 3 × 2 grid you should already know. Export 9:16 1080p as the workhorse for Shorts and in-feed. Add 1:1 and 16:9 derived from the same master so Discover, Gmail and in-stream-adjacent surfaces have a file that is not letterboxed. Headlines should match the hook family, not a generic brand slogan that none of the videos say, or the text and the picture will argue in the same auction.

Label every asset with hook and avatar codes before upload so the asset report is readable without downloading files. Google will mix images and videos; if you only care about UGC video, still add a few product images rather than leaving those slots empty, then watch which formats actually spend. Keep claims consistent with the landing page. A Demand Gen asset group is a public ad across YouTube and Discover, not a thumbnail experiment you can treat as informal. If a headline overclaims, the video should not be allowed to carry it.

Demand Gen uses audience signals to start: customer match lists, site visitors, YouTube viewers, custom segments from search terms or apps, and lookalikes (Google's similar audiences / lookalike-style expansion on those seeds). Upload the best first-party list you have. Do not stack twenty interests until delivery dies. The system is supposed to expand; your job is a relevant seed, not a replica of a 2018 display plan.

Customer match is the strongest seed if the list is large and recently hashed from a real CRM export, not a leftover spreadsheet. A remarketing list of purchasers or high-intent visitors is next. Custom segments built from competitor and category search terms can stand in when first-party data is thin. Lookalikes of purchasers are the usual prospecting move once the source list qualifies. If a list is tiny, skip the lookalike and use a broader custom segment rather than pretending 80 emails are a twin audience the system can model. Garbage seeds make expansion look random.

Use these as signals, then let expansion work. Tight remarketing-only Demand Gen is a different job — useful for warm traffic, misleading if you thought you were prospecting and then compared CPA to a cold Meta campaign. Exclude converters if the campaign should find new buyers. Check that list membership is updating; a frozen 2023 customer file is a stale hint that quietly ages. Audience setup should take less time than creative. If it takes longer, you are over-targeting, and you will starve the asset group you just spent a sitting building.

Start with Maximize conversions if you lack a stable CPA, then move to Target CPA once you have a recent history of conversions to aim at. Set the initial target a little looser than your true goal so the campaign can buy volume and learn. A Target CPA that is far below what the account has ever paid usually starves delivery and produces a 'Demand Gen does not spend' ticket.

As a rough operating rule, automated CPA bidding wants a run of conversions in a recent window before it feels stable — think on the order of a few dozen in a month, not two purchases and a hope. Below that, expect swingy daily CPA and do not retarget the bid every morning because yesterday looked expensive. Budget should be able to buy those conversions at a plausible cost; a tiny daily budget with a strict Target CPA is how Demand Gen sits at zero impressions while you refresh the graph. Fund the learning or do not launch yet.

Target ROAS is for accounts with reliable conversion values coming through, not for a first test with messy revenue import or missing transaction IDs. If values are missing, stay on conversions until the numbers are trustworthy. Give the bid strategy a week of mostly untouched delivery. Large budget cuts, strategy flips and asset-group rewrites all reset learning. If you must change something, change one thing, then wait again. The auction is slower to educate than your inbox, and it does not care that you refreshed twice.

Publish, then leave the campaign alone aside from policy and tracking checks. Watch whether Shorts, in-feed and other surfaces are actually serving — open the placement or inventory view rather than assuming the name Demand Gen did the targeting for you. Early CPA is often ugly. Ugly plus converting events is still a learning campaign. Zero conversions plus you rewriting headlines on day two is a self-inflicted stall.

If spend dumps into Display or Gmail and you wanted Shorts, check asset mix and inventory controls rather than killing the whole campaign on a first-look CPM. If Shorts serves but view metrics look 'worse' than in-stream, remember the environments count views differently — judge on conversions and CPA, with view rate as a diagnostic, not a kill switch. If one asset takes all the spend, that can be fine if it converts; it is a problem if it is the leftover 16:9 file beating better vertical UGC only because the system found cheap inventory for letterboxed video.

Do not add a second asset group on day three 'to help learning' or because a stakeholder wanted another headline. Finish one clean group. Frequency and creative fatigue come later, after you have a baseline. First you need a CPA you could explain and a sense of which hook the asset report likes. That takes days, not hours. Put a calendar reminder at day seven so you are not hovering, and put the next 3 × 2 grid on the calendar only after that readout exists in writing for the team to act on next.

At the first honest readout, sort videos by conversions and cost per conversion, not by view rate. Keep the hook family that wins, kill the files that spent without result, and generate the next 3 × 2 grid from that insight. Scale budget in modest steps. Duplicate into a new campaign only when you have a reason — inventory split, budget cap, a new offer — not as a superstition.

Asset-level reporting is the point of loading variation in the first place. If you uploaded six near-identical cuts, the report cannot teach you a hook, only which filename the system happened to spend on. Write the winning opening line down and make the next batch around that angle with new faces or a new offer, one variable at a time. Export new 9:16 1080p files rather than stretching the winner into a 60-second cut and hoping Shorts holds. Treat a 30%+ hook-rate floor as a health check once views exist, then still rank on the conversion you pay for.

Klip Kanvas is the production side of this loop: product link to a 3 × 2 UGC set, 3–5 minutes per 1080p render, 9:16 plus 1:1 and 16:9 from the same project so the asset group is not a graveyard of one ratio. Google still owns the auction. Your job is enough distinct assets, a seed audience, a bid you can fund, and the patience to read conversions before you rebuild the campaign around a single day's CPM. When frequency climbs, refresh the hook; do not only raise budget on a tired file and hope the auction forgets.