There is an email going around to agencies right now offering up to $1,500 in Google Ads credit to put clients into Demand Gen campaigns. It comes with a copy-and-paste pitch email, a blank space for the client's name, and a shortlist of which of your accounts are eligible.
It is a good offer. It is also the reason to think carefully rather than quickly, because a credit removes the one thing that normally forces a decision to be examined, which is whether it is worth the money.
So here is the whole picture. What Demand Gen is, what it is genuinely good at, the three situations where it earns its place, the situations where it will quietly waste a quarter, what it actually costs, what assets it demands, how to set it up so it has a chance, and the measurement problem that sits underneath all of it.
That last part is the one almost nobody covers, and it is the part that decides whether any of the rest matters.
What Demand Gen Actually Is
Demand Gen is Google's answer to paid social. It runs your images and videos across YouTube, including in-feed placements, Shorts and in-stream, plus Discover, Gmail, Maps and the Google Display Network.
It began life as Discovery campaigns. It has since absorbed them, and it is now absorbing Display too. Google's documentation says eligible advertisers could begin migrating Display campaigns from June 2026, and that new campaigns will eventually only be creatable inside Demand Gen, with remaining eligible campaigns migrated automatically. We covered what that migration takes away in Google Calls It a New Home for Display. Every Feature It Removed Was a Manual One.
The important part is not the surfaces. It is what Google says the campaign is for. In its own FAQ, Demand Gen "uses AI to serve your best-performing visuals to audiences who are most likely to be interested in your brand, even if they aren't actively searching for you yet."
Read the last nine words again. That is the product's definition of itself, and it is the honest one.
The Structural Fact That Explains Everything Else
Google publishes a comparison table for Search, Performance Max and Demand Gen. It is worth reading closely, because one row settles most arguments about this campaign type.
Under channels, Search reaches Search and search partners. Performance Max reaches everything, Search included. Demand Gen reaches YouTube Shorts, YouTube in-feed, YouTube home feed, YouTube watch next, YouTube search, Discover, Gmail and the Display Network.
There is no Google Search text inventory in Demand Gen. That is not an oversight and it is not a gap Google intends to close. It is the definition of the product. Search campaigns capture demand that already exists. Demand Gen goes and creates some.
One honest caveat, because this argument is often stated too strongly. Google added Maps inventory in June 2026, and Maps carries genuine local intent. So the clean line that Demand Gen "cannot touch intent" is no longer quite true. What remains true is that it has no access to the query, which is the thing that makes Search legible to a small advertiser.
The Three Times Demand Gen Is the Right Call
Aaron Young of Define Digital Academy, who is among the more careful practitioners publishing on this, names three scenarios. They are the most useful framing available, and notice what all three have in common.
You have hit CPC resistance. Your Search campaigns will not scale any further. You raise the budget by ten percent and your cost per click jumps twenty-five or thirty percent, so growth stops being profitable. Optimizations inside Search are no longer moving anything. At that point the ceiling is the auction, not your account, and you need a different room.
Your product needs explaining. You sell something materially similar to competitors but better on a dimension nobody searches for, or you serve a niche, or you are in education, coaching or a considered-purchase category. Nobody types a query for a benefit they do not know exists. Demand Gen can carry that story where a text ad cannot.
Your Meta account has plateaued. You are already running paid social successfully, you have creative that demonstrably converts, and returns are diminishing. Porting proven assets onto YouTube and Discover is a genuinely sensible move, and it is probably the single most reliable Demand Gen use case there is.
Here is the pattern. Every one of those three presupposes that something else is already working. CPC resistance means you have a Search campaign at its ceiling. Diminishing Meta returns means you have a Meta account worth diminishing. Demand Gen is a scaling channel. It is not a starting channel, and Young says as much himself: most businesses will get where they are going quicker by pushing harder on the campaign types they already run before adding this one.
When It Is the Wrong Call
Before your demand capture is finished. If there are searches for what you sell that you are not showing up for, buying awareness is buying the harder half of the problem first. Capture is cheaper than creation, always. This is the same argument we made about the campaign menu generally in You Have $2,500 a Month. Google Offers You Eight Campaign Types.
When you have no creative pipeline. Not "no creative." No pipeline. Demand Gen consumes assets continuously and needs refreshing. Young puts this plainly as one of his five biggest mistakes: you have to think about not just the ad spend but whether you have the budget and the team to keep producing new creative, and if you do not, do not start.
When you cannot measure independently. This is the disqualifier almost nobody applies, and the rest of this article is about why.
When the budget is below the floor. Which is a specific number, and it is higher than you think.
What It Actually Costs
Google now enforces a hard minimum. Its developer blog announced that from April 1, 2026, the Google Ads API began enforcing a minimum daily budget of five US dollars, or local equivalent, for Demand Gen campaigns.
Five dollars is a floor, not a recommendation. Google's own setup guidance for Demand Gen says that for campaigns using target CPA bidding, it recommends a budget at least ten times your target CPA. If a lead is worth a $150 target CPA to you, Google is describing a $1,500 daily budget.
You will see higher numbers quoted. Young says twice that Google recommends $100 a day or twenty times your cost per conversion. We could not verify a twenty-times figure in Google's published documentation, which states ten times, so treat the higher multiples as practitioner caution rather than Google policy. Either way the direction is the same and it is not five dollars.
Young's own worked example is the clearest illustration of the failure. A campaign running $10 a day, targeting the entire United States, on YouTube only. He puts the US YouTube audience at roughly 250 million people. Whatever the exact figure, ten dollars against a market that size is not a small budget, it is a rounding error, and it will teach you nothing.
There is genuine nuance on the other side, and he supplies it: he has seen good results at $30 to $50 a day, but only when the targeting was tightly constrained to smaller cities, specific segments, or audiences built from existing conversion data. The variable that matters is not the budget alone. It is the budget relative to the size of the market you pointed it at.
And then time. Young's position is that you give a Demand Gen test at least three months before judging it, because you are reaching a colder audience on surfaces where the gap between seeing an ad and acting on it is long.
The Measurement Problem, Which Is the Real Story
Here is where this article stops agreeing with most of what is written about Demand Gen, in both directions.
The common critique is that Demand Gen inflates its results. The common defense is that attribution models unfairly starve it of credit. The interesting thing is that both camps have documented, credible evidence, and they are describing the same underlying defect.
The case for over-reporting
In February 2025, John Moran published a teardown of a Demand Gen campaign that had spent $30,000 in about twenty days. The headline numbers looked superb: conversions rose from 137 to 377 and eventually past 1,500, and a nine percent increase in cost appeared to be delivering eighty percent more conversions.
Splitting the data by ad event type told a different story. Engaged View Conversions, which are counted when someone watches a YouTube ad for at least ten seconds and converts within a window without ever clicking, went from 205 to 1,334 in two weeks. Click-based conversions rose only modestly. Qualified leads in the CRM barely moved at all, and the UTM parameters that should have accompanied hundreds of new leads were not there.
The detail that settles it: brand-search conversions fell by fourteen percent during the same period, while the brand campaign was capturing roughly ninety-four percent click share. Demand had not increased. Credit had moved.
His conclusion was that the campaign was "predominantly scooping up EVC credit for people who already knew about the brand or were already going to convert." Worth noting that he is not hostile to the product. He says explicitly that Demand Gen is not inherently flawed and can be a powerful prospecting channel. The problem was measurement, not the campaign type.
The case for under-reporting
Now the opposite, from Young, with equally specific numbers.
A client campaign had spent roughly $3,000 and Google Ads reported fourteen conversions. On that evidence you would switch it off. But the business had placed its own tag in its database, recording when a signup first arrived from the Demand Gen campaign. Followed through, those signups had produced about $8,500 in revenue within thirty days, and against a known lifetime value of roughly $990 per signup, more than $15,000 in total revenue from that $3,000.
He also describes a cleaner proof: an Australian business ran Demand Gen in a single capital city only, so the effect could be isolated geographically. Purchases in that city rose more than thirty-five percent and cost per conversion fell from $120 to under $70 within six weeks.
His summary of the whole problem is the most useful sentence anyone has written about this campaign type: it either over-reports or it under-reports the effectiveness of Demand Gen.
Why both are true
These are not competing claims. They are two symptoms of one condition. Demand Gen operates where the connection between the ad and the action is loose: no query, often no click, a long gap between exposure and purchase, and heavy overlap with audiences who already know you.
When that campaign runs alongside a strong brand presence with no exclusions, it absorbs credit it did not earn. When it genuinely creates demand that converts later through another channel, it gets no credit for work it did do. Same mechanism, opposite signs, and no way to tell which one you are looking at from inside the interface.
Which produces the rule this whole article has been building toward:
The number Demand Gen reports is not evidence. Treat it as a hypothesis and go and test it somewhere else.
That is a manageable requirement for a business with a CRM, clean UTMs and someone who can run a geo holdout. It is a genuinely hard requirement for a small business with a shared inbox and a phone. Which is precisely the group being offered credits to try it.
How to Actually Know Whether It Worked
Four methods, cheapest first. You do not need all of them, but you need more than the conversions column.
Segment by ad event type. In the campaign, split conversions into click-based versus engaged view. If engaged-view conversions climb while click conversions stay flat, you are looking at Moran's pattern and you should assume credit reassignment until proven otherwise.
Watch the channels Demand Gen could be stealing from. Brand search conversions, direct traffic, organic. If in-platform conversions leap while those fall or stay flat, total demand did not grow. If they rise together, something real is happening.
Match against your own records. Count leads in your CRM, not conversions in Google Ads, and check that UTM parameters actually arrive on the forms. Moran's tell was that the CRM and the UTMs did not corroborate the platform. Young's tell was a database tag that proved the opposite.
Run a geo holdout. The strongest method and the least used. Run Demand Gen in one city or region and not in a comparable one, then compare total business outcomes between them, not platform metrics. This is what isolated the Australian result, and it is the only approach on this list that survives contact with a skeptical finance person.
Set exclusions before any of this. Moran's specific recommendation is to exclude website visitors from the last ninety days or longer, plus existing customers, so the campaign cannot quietly retarget people who were already going to convert. Without exclusions you are not testing demand creation. You are paying to re-buy an audience you already had.
The Assets It Demands
This is where most small advertisers discover the real price.
Google's asset specification for Demand Gen covers images in landscape 1.91:1, square 1:1, portrait 4:5, plus a 1:1 logo; text including headline, description, final URL, business name and call to action; video in landscape, portrait, square and vertical; and optional carousel ads carrying between two and ten image cards. Ecommerce advertisers can attach a Merchant Center product feed.
You can run image-only or video-only. Google's FAQ confirms both are allowed while noting that using both typically performs better.
On top of the formats, Young recommends three or four genuinely different creative angles per ad group, not three crops of one idea, so the campaign has something to learn from. And videos must be uploaded to YouTube before you can use them, which catches people out. Uploading them as unlisted keeps them out of your organic channel.
Add that up honestly before starting. Four image ratios, four video orientations, several angles each, refreshed on a cycle. For a business without a designer or an editor, that is the actual barrier, and no amount of ad credit removes it.
How to Set It Up So It Has a Chance
If you have decided it fits, these are the settings that separate a fair test from a wasted quarter.
Build ad groups around placements, not audiences. One for Shorts, one for in-stream, one for Discover or Display. Young's reasoning is practical: performance differs enormously by placement, and if they are mixed in one ad group you cannot see which is working or pause the one that is not.
Pick one conversion action for bidding. Not everything in the account. Choose the action that represents real business value and remove the rest from this campaign's optimization.
Be careful what Google tells you to optimize for. Google's own setup guidance recommends optimizing for "lightweight conversion events such as 'add to basket' or 'site visit'." For an ecommerce store finding signal, that is defensible. For a lead-generation business it will train the campaign toward browsers, and you will get a healthy-looking number made of people who will never call. We wrote about that failure mode generally in Your Conversion Number Is Going Up. Your Sales Aren't.
Do not set a target CPA at launch. You do not yet know what a conversion costs on these surfaces, and guessing low starves the campaign before it can learn. Add the target once it has data.
Turn optimized targeting off if your targeting is deliberate. This is Young's third mistake, and it is a good one. Optimized targeting lets Google go beyond the audiences and demographics you selected whenever it thinks it can find more conversions. If you carefully built an audience of cart abandoners, leaving optimized targeting on can quietly undo that work. Leave it on only when you genuinely want reach.
Set location targeting to "presence" only. Not presence or interest, which will serve your ads to people merely reading about your city.
Mind the audience minimums and the exclusions. A custom audience needs at least 100 users before it can be used. Exclude existing purchasers. Lookalike segments built from your customer list are one of the genuinely strong targeting tools here, and they are a real advantage over Performance Max, which offers no hard targeting at all. We compared that control gap in Performance Max Isn't Your Smartest Campaign. It's Google's.
Know what you gave up. Google's FAQ confirms that core brand-suitability settings are not available at campaign level, and that topic and placement exclusions are not generally available at campaign or ad group level. Content exclusions are managed at account level. If placement control matters to you, that is the trade.
Does It Work for Lead Generation?
It can, and the fair answer requires citing the strongest evidence against skepticism.
Google's own case study, published September 2025, features ServiceTitan, a home and commercial services software company. They ran two Demand Gen strategies, one mid-funnel for subscriptions and one down-funnel for marketing qualified leads, and used Lookalike segments built from existing customers. Google reports a thirty-six percent increase in conversion volume and a seven percent reduction in cost per acquisition. Against their previous paid social campaigns: 307 percent more conversions, a ninety-four percent higher conversion rate from initial lead to booked meeting, and roughly half the cost per subscriber.
Those are strong numbers and there is no reason to dismiss them. But read what they actually describe. A large software company, with existing customer lists rich enough to build Lookalikes from, with creative already made for social, benchmarked against its own previous social campaigns rather than against starting from nothing. That is Young's third scenario almost exactly: an established advertiser hitting a ceiling on social and porting proven creative to Google.
It is a compelling argument for Demand Gen as a scaling channel for a business that has already solved creative and measurement. It is not evidence that a plumber with a $2,500 monthly budget should be pointed at YouTube.
So What Do You Do With the Credits?
If someone offers you $1,500 to try Demand Gen, the offer is real and the money is real. The question is whether you can answer one thing when it ends: did this create demand that would not otherwise have existed?
If you can run a geo holdout, or tag leads into a CRM, or you have enough brand search volume to notice it moving, then yes, take the credit and run a proper test with exclusions in place and three months on the clock.
If you cannot, the credit does not solve your problem. It pays for the ad spend and leaves you with the harder bill, which is creative production, and an answer you will not be able to trust. A free test that produces an unreliable result is not free. It costs you the decision you make next.
The Honest Verdict
Demand Gen is a competent, improving product with more control than Performance Max: you choose the placements, you can target Lookalike segments, you get per-asset reporting and real creative experiments. The people calling it a black box are not describing it accurately.
It is also structurally a demand-creation channel with no access to the search query, which means it should be bought after demand capture is finished rather than instead of it, and it should be measured outside the platform because inside the platform it is unreliable in both directions.
The businesses that do well with it look alike. They have a working Search account at its ceiling, creative they can keep producing, a defined audience, a budget matched to the market they aimed at, three months of patience, and some way of counting outcomes that does not come from Google.
If that describes you, the credit is a genuine opportunity and this is a good time to take it. If it does not, the honest move is to keep building the thing that is already working and revisit this when the ceiling arrives. We spend a lot of our time in Google Ads management telling people the second thing, and it is rarely what anyone hoped to hear.
Either way, the discipline is the same. Do not let a campaign grade its own homework.
