Digital Advertising Playbook

Digital advertising rewards discipline more than budget. The platforms’ bidding learns from whatever you feed it, so the work is to give it a clear job, a true conversion and clean inputs, then to judge each change only when the numbers can tell you something.

This playbook is about running the campaigns, platform by platform: Google, Meta and LinkedIn in the examples, but the steps apply to any platform that bids for you. How much to spend, and how to split it across channels, is in the Paid Media Budget Allocation Playbook. Which channels to test first, and how to tell whether the ads caused the sales they claim, is in the Demand Gen & Performance Marketing Playbook.

Every campaign should move one group of people one stage closer to buying, and you should be able to say which stage in a sentence. Four jobs cover almost everything:

• Awareness: build reach and familiarity among people who do not know you. Measure reach, video views and, where the platform offers a brand-lift study, ad recall.
• Consideration: move people from awareness to active interest. Measure visits to key pages, content downloads and add-to-carts.
• Conversion: turn interest into revenue or qualified leads. Measure purchases, trial starts, demo bookings and qualified leads.
• Retention: earn more from existing customers. Measure renewals, upsells and reactivations.

The platforms ask the same question in their own words. Meta folded its eleven campaign objectives into six, Awareness, Traffic, Engagement, Leads, App promotion and Sales^[1], and LinkedIn groups its objectives under awareness, consideration and conversion^[2]. The objective tells the platform whom to find, so choose it from the job, not from the cheapest result: in my view, a campaign optimized for clicks finds people who click.

Channels follow the job. Search, Shopping and Performance Max catch people already looking; YouTube, Meta and TikTok reach people before they look; remarketing and customer lists work on people who already know you. Google’s old Discovery campaigns are now Demand Gen campaigns, which run on YouTube, including Shorts, Discover and Gmail^[3]. Which channels to fund, and with how much, is Steps 2 and 5 of the Paid Media Budget Allocation Playbook.

Every bidding system learns from the conversions you report, so measurement comes before launch, not after. It has three layers:

• Tags in the browser: the Google tag or Google Tag Manager, the Meta Pixel, the LinkedIn Insight Tag and the TikTok Pixel, each firing on the same events.
• The same events from your server: browser tags miss conversions when cookies are blocked or a page closes early. Meta recommends its Conversions API alongside the Pixel, and counts the two once when both send the same event name and event ID within 48 hours^[4]. Google’s enhanced conversions sends the email from your form, hashed with SHA-256 before it leaves the page, to match conversions to signed-in Google users^[5].
• Names that match: use each platform’s standard event names. GA4’s include view_item, add_to_cart, begin_checkout, purchase, sign_up and generate_lead ^[6]; Meta’s include ViewContent, AddToCart, InitiateCheckout, Purchase, Lead and StartTrial ^[7].

Then decide which conversion the bidding chases. In Google Ads, only primary conversion actions appear in the Conversions column and steer bidding; secondary actions are reported under All conversions, for observation^[8]. Make the action closest to revenue that still happens often enough your primary, and everything else secondary; how often is enough is Step 6 of the Paid Media Budget Allocation Playbook. Sending later outcomes, such as a closed deal or a paid subscription, back from your CRM is Step 5 of the Demand Gen & Performance Marketing Playbook.

Tag every ad link with UTM parameters, and make utm_campaign the campaign’s name from Step 3, so GA4 and your CRM read the same labels; the GA4 Tracking Plan and UTM Builder keeps event names and UTM rules on one sheet. If you advertise to people in the European Economic Area, Google also needs their consent: consent mode v2 adds the ad_user_data and ad_personalization signals, and you must collect consent from EEA users and pass it on^[9]. The privacy and advertising rules post covers the wider rules.

Structure has two jobs: give each campaign enough conversions to learn from, and make reports readable without a spreadsheet. The levels work the same way on every platform:

• Campaign: one job from Step 1, one objective, one budget and one bid strategy. Split campaigns only where the job, the market or the budget really differs.
• Ad set or ad group: one audience idea, such as one keyword theme or one remarketing list, with its placements and schedule.
• Ad: several variations per ad group, in different formats and angles, so the platform can find what works (Step 5).

The common mistake is too many small pieces. Every extra ad set splits the same conversions more thinly, and one that collects too few stays in learning; Step 6 of the Paid Media Budget Allocation Playbook gives each platform’s threshold. In my view, a startup spending under about $20,000 a month rarely needs more than two or three campaigns per platform. I also keep brand searches in a campaign of their own: people typing your name convert cheaply, and mixed in with everything else they make non-brand search look better than it is.

Name everything to one pattern, so the name says the platform, the job, the audience and the market, and reuse it as utm_campaign.

Audiences come in three layers, from strangers to customers, and each gets its own message (Step 5):

• Prospecting: people who do not know you. On search, the keyword is the audience. On social and video, start broad and let the bidding find buyers within it. Google stopped offering similar audiences in August 2023 and points advertisers to optimized targeting, audience expansion and Smart Bidding instead^[10].
• Remarketing: people who visited, watched or engaged, such as pricing-page visitors, people who started a form, and video viewers. Warmer and cheaper per conversion, but capped by your traffic.
• Customers: lists uploaded with consent and used three ways: excluded from acquisition, targeted for upsell, and used as a seed so the platform can find similar people. A Meta lookalike needs a seed of at least 100 people from one country^[11].

Then write the exclusions, because that is where budget leaks. Customers come out of prospecting and remarketing; people already in a trial come out of trial ads. Meta removed detailed targeting exclusions, such as excluding an interest, from July 2024^[12], but custom audiences can still be excluded, and since August 2026 Meta offers exclusion-only custom audiences that cannot be used for targeting by mistake^[13].

Check sizes before you spend. Google serves a remarketing or customer list on Search, Display and YouTube once it has at least 100 active users in the last 30 days^[14]; LinkedIn will not run a campaign whose audience has fewer than 300 members^[15]. The Paid Audience Spec Builder checks each audience against its platform’s minimum.

Some categories narrow targeting by policy. On Meta, ads for housing, employment, financial products and services, or social issues, elections and politics must be declared as a special ad category; for the first three, targeting by age, gender or ZIP code and lookalike audiences are not available^[16]. Google bars age, gender, parental status, marital status and ZIP-code targeting for housing, employment and consumer finance ads in the US and Canada^[17]. A fintech, HR or property startup has to plan its audiences inside those limits from the first campaign.

Creative is the lever you control most. In NCSolutions’ 2023 update of its study of about 450 consumer packaged goods campaigns, creative quality drove 49% of the incremental sales, more than brand, reach, targeting or recency^[18]. That evidence comes from groceries, not software, so treat the number as a direction rather than a rule.

Match the message to how warm the audience is:

• Cold: lead with the problem or the outcome, in the customer’s words, and ask for little: a guide, a short video, a quiz.
• Warm: show why you are the better answer, with a demo, a comparison, proof from customers and the main objection answered.
• Ready to buy: make the offer and remove the risk, with a free trial, a guarantee, clear pricing and one call to action.
• Customers: help them get more from what they bought, and offer the next product when it fits.

Inside an ad the order rarely changes: a hook in the first line or first seconds, the problem, how you solve it, proof, and one call to action. Offers follow the same temperature: low-commitment offers for cold audiences, stronger ones with a deadline and proof for warm audiences, and always the same promise on the landing page (Step 6).

Format rules worth knowing

• Google search ads are responsive: you write up to 15 headlines of 30 characters and up to 4 descriptions of 90 characters, and Google assembles them. Put the main keyword in at least one headline. Pinning a line to a position “isn’t recommended for most advertisers”, so pin only what must always show, such as a disclaimer^[19].
• YouTube viewers can skip an in-stream ad after 5 seconds, and a Shorts ad at once^[20]. Google’s ABCD principles, attention, branding, connection and direction, say to show the brand early and throughout and to back up the audio with on-screen text; Google cites Google and Kantar work linking them with as much as a 30% lift in short-term sales likelihood^[21].
• Reels, Stories, Shorts and TikTok are full-screen vertical feeds: shoot vertical, and caption every line so the ad makes sense with the sound off.

Every claim needs evidence you could show a regulator. Under the US Federal Trade Commission’s rule announced in August 2024, fake or AI-generated reviews, reviews bought for a set sentiment and undisclosed reviews from staff can bring civil penalties^[22]. The Marketing Compliance Checklist covers the check before launch.

An ad can only be as good as the page it sends people to. Four rules:

• Message match: the headline repeats the ad’s promise in the same words, and the ad’s offer is the page’s offer.
• One goal: one primary action per page, such as a purchase, a sign-up or a booking.
• Little friction: fast on a phone, with only the form fields you need now.
• Reassurance: reviews, security notes, clear pricing and short answers to the obvious questions.

Google measures part of this. Quality Score, a 1 to 10 diagnostic for search keywords, rates landing page experience alongside expected click-through rate and ad relevance; it is not an input to the auction, but its parts show what to fix^[23]. Speed matters too: in a study commissioned by Google from 55 and Deloitte, which tracked 37 brand sites at the end of 2019, a 0.1-second improvement in four speed metrics went with 8.4% more retail conversions and 21.6% more lead-generation visitors reaching the form page^[24]. For a yardstick, the median US search campaign turned 8.18% of clicks into a lead or sale from April 2025 to March 2026, and 4.85% in business services^[25]. The Website & Conversion Optimization Playbook covers page design in depth.

Automation now does most of the work: bidding, placements, audience expansion, even ad text. Use it, but decide what it may change, and check what it did.

Bidding. Google’s advice follows your goal: Maximize conversions, with an optional target cost per action, when you want conversions; Maximize conversion value, with an optional target return on ad spend, when conversions differ in value; and Maximize clicks for a new campaign without enough conversion data^[26]. Check first that the primary conversion is the right one (Step 2), because Smart Bidding chases whatever you count. On Meta, a sales, app or leads campaign counts as Advantage+ only when its budget, audience and placements are all left to the system^[27].

Broader matching and generated ads. Google’s AI Max for Search, launched in May 2025, extends keywords to related searches, writes extra headlines from your site and can send people to other pages on it. Google says advertisers typically see 14% more conversions at a similar cost, a figure from its own data^[28]. Those extra conversions come from searches and pages you did not choose, so read the search terms and the generated text before you trust them. To price a keyword before you bid, see estimating what a keyword will cost in the Paid Media Budget Allocation Playbook.

Guardrails. The controls that keep automation inside your business:

• Negative keywords and brand exclusions. Performance Max gained campaign-level negative keywords in 2025^[29], and the limit rose from 100 to 10,000^[30].
• Auto-applied recommendations, off or reviewed. Google can apply recommendations on a schedule, including adding broad-match keywords and changing bid strategies or targets^[31]. Turn on only the ones you would accept by hand.
• Brand safety settings. Google sets content suitability for video and display ads at account level, with Maximum, Moderate and Limited inventory^[32]. Add exclusion lists for sites and channels you never want to appear on.
• A person checks anything generated. AI can draft hooks, headlines and scripts, but someone checks every claim against Step 5 before it runs.
• Your own numbers as the judge. Platforms report the conversions they can see; judge automation on what reaches your CRM or billing, such as paying customers.

Test in the order that usually moves results most. This is my order, from experience rather than a study:

1. The offer and value proposition: the core promise.
2. The creative concept: the story, angle or hook.
3. The audience.
4. Bidding and budget settings.
5. The landing page.
6. Small details: button text, colours and placement.

Each test needs a hypothesis, one success measure, such as cost per acquisition or conversion rate, and enough data to read. How much depends on your conversion rate and the size of change you want to see. A standard approximation is 16 × p × (1 − p) ÷ d² people per version, for 95% confidence and 80% power, where p is the current rate and d the change^[33]. At a 4% conversion rate, seeing a rise to 5% takes about 6,144 visitors per version. Small changes need traffic most startups do not have, so test big ones. Inside Google Ads, Google suggests running an experiment for at least four to six weeks, or one to two conversion cycles, and leaves its first seven days out of the results^[34]. The A/B Test Significance Calculator reads the result, and the Paid Media Test Log keeps each test’s hypothesis and read date on record.

Then hold one rhythm:

• Daily, a few minutes: spend against plan, disapproved ads, broken tracking and sudden swings.
• Weekly: search terms (estimating what a keyword will cost shows how to read them after a test), tired creative, new variants and at most one change per test; Step 6 of the Demand Gen & Performance Marketing Playbook sets out the review.
• Monthly: results by channel, stage, audience, creative concept and offer, judged on customers rather than leads, and the budget moves that follow, which are Step 8 of the Paid Media Budget Allocation Playbook.

Each report should answer what its review decides: spend, conversions and cost per conversion by channel; drop-off from click to lead to customer; results by creative concept and format; and, once you have a few months, how customers from each channel and offer behave over time.

What changes by stage?

What mistakes should I avoid?

• Optimizing for the cheapest event. Bidding finds more of whatever you count; make the event closest to revenue your primary conversion (Step 2).
• Counting one conversion twice. A Pixel and a Conversions API without a shared event ID inflate every result (Step 2).
• Splitting the account into pieces too small to learn. Fewer, fuller campaigns learn faster (Step 3).
• Showing acquisition ads to customers. Exclude them everywhere (Step 4).
• Sending every ad to the homepage. Each ad needs a page that repeats its promise (Step 6).
• Calling a test after a week. Read it on the date its sample size sets (Step 8).