An ad says Start your free trial. The click leads to a form asking for a work email, company size, phone number, and a description of the project. The success screen says someone from sales will follow up.
The problem is not that Start is a weak verb. The problem is that no trial starts.
A useful ad call to action works like a small contract. It connects four things: the question the customer is trying to answer, the action the ad promises, the conditions attached to that action, and the result the destination can actually deliver. When those four parts disagree, polishing the button usually leaves the real problem intact.
That does not mean you can look at an ad and diagnose “the customer’s biggest doubt.” A question can come from search behavior, sales calls, support tickets, user interviews, or observed tasks. Without that evidence, it is a plausible hypothesis—not measured customer psychology. The practical goal is simpler: choose a question worth answering, then make the ad and destination answer it honestly.
A CTA should name the job the next screen can finish
“Learn More,” “Get Started,” and “Shop Now” are not inherently good or bad. Their value depends on what happens after the click.
The distinction matters because people may click an attractive general CTA while looking for a very specific answer. In a 2017 usability article, Nielsen Norman Group described participants who clicked prominent Get Started buttons while trying to understand a service, only to enter signup flows before getting the information they expected. The study was not an ad-wording A/B test, but it illustrates the operational problem: a click is not the same as completing the visitor’s task. See “Get Started” Stops Users.
Before writing the CTA, fill in this six-part contract:
| Contract field | What to write | Example |
|---|---|---|
| Customer question | A researched question, or a hypothesis labeled as such | “Will this dock work with my hardware revision?” |
| Ad promise | The specific next action | “Check your exact model” |
| Conditions | Price basis, account, form fields, region, timing, or other gates | “Model and revision required; no email” |
| Destination | The page, native form, product state, or tool that receives the click | Compatibility checker |
| Available result | What the person can actually get, including negative outcomes | Compatible, incompatible, or not yet verified |
| Completion event | The event that proves the promised job was finished | compatibility_result_viewed |
The question should be grounded when possible. The GOV.UK Service Manual’s user-research guidance recommends using existing research, analytics, search and support records, interviews, and observation rather than treating internal opinions as user evidence. That guidance comes from public-service design, not ad conversion testing, but the boundary transfers cleanly: evidence can support the question; a team hunch should remain a hypothesis.
12 ad call-to-action examples
The twelve examples below are original fictional teaching cases. Their businesses, prices, products, URLs, and interface states are invented; none is a tested campaign or a claim about conversion lift. Each one shows a bad-to-better change across the ad, CTA, and destination—not just a stronger-sounding verb.
1. Let a price question lead to a real price
Customer question: “Can I see what the plan costs without speaking to sales?”
Mismatch: Start your free trial leads to a paid plan page, even though no trial exists.
Better ad: “One reporting plan: $30 per account each month, including five users and 200 reports. Check the price and limits before you sign up.”
CTA: See the monthly plan
The destination should display “$30 per account per month,” monthly billing, the five-user and 200-report limits, no setup fee, and the fact that tax may be extra—without requiring an account to reveal those terms. Twelve separate monthly payments would total 12 × $30 = $360 before tax, but that is not an annual-prepayment price or an annual discount. Completion is viewing the price-and-limits card; purchase is a separate event.
2. Distinguish an estimate from a final quote
Customer question: “What would assembling these two items roughly cost?”
Mismatch: Get your final quote instantly opens either a rough calculator or a contact form.
Better ad: “Estimate standard furniture assembly in ZIP codes 10001 and 10002. Choose your items to see the visit fee and item prices before requesting a quote.”
CTA: Calculate your estimate
For one desk and one bookcase in ZIP 10001, the fictional calculator shows a $30 visit fee, $90 desk assembly, and $60 bookcase assembly: 30 + 90 + 60 = $180 before tax. It also states that wall mounting and repairs are excluded and that the result is not a confirmed quote. ZIP 94103 should produce “We do not currently serve this ZIP code. No estimate has been generated,” not a fake $0 result. When the business cannot calculate a price, change the promise to Request an assembly quote and disclose the response process instead.
3. Treat an incompatible result as successful fulfillment
Customer question: “Will this dock work with my exact model and revision?”
Mismatch: Works with every H100 leads to a page where H100 Rev A is supported but H100 Rev B is not.
Better ad: “Check your device model and revision before ordering. Our checker distinguishes supported, unsupported, and unverified models.”
CTA: Check your model
The destination returns three honest states: H100 Rev A is compatible; H100 Rev B is not compatible and should not be ordered; H200 has not been verified. The checker should not turn “unknown” into “yes,” and it should not require a lead form to reveal the result. A negative answer can be the correct completion because the promised job is to check compatibility, not to force every visitor toward checkout.
4. Replace “perfect fit” with the measurements you actually have
Customer question: “How does this shirt compare with one that already fits me?”
Mismatch: Find your perfect fit leads to a generic S/M/L chart that does not say whether the figures are body or garment measurements.
Better ad: “Compare our shirt’s garment measurements with one you already own. See chest measurements in inches and centimeters before choosing a size.”
CTA: View garment measurements
The destination labels the numbers precisely: S is 40 inches around the garment chest, M is 42, and L is 44; the corresponding flat half-chest widths are 20, 21, and 22 inches. For M, 42 × 2.54 = 106.68, rounded to 106.7 cm. A shopper who prefers four inches of ease over a 38-inch body chest might compare with the 42-inch garment, but that preference is not a fit guarantee. The CTA promises usable measurements, not certainty the page cannot provide.
5. Show the proof artifact without inventing the outcome
Customer question: “What would the report look like, and how is its total calculated?”
Mismatch: See how customers doubled revenue leads to a fabricated or purely illustrative report with no customer study behind it.
Better ad: “See a worked sales report with its source rows and calculation. The example uses fictional data, not a customer performance claim.”
CTA: See the sample report
The sample contains two days of fictional data: Monday has $600 in gross item sales and $50 in refunds; Tuesday has $400 and no refunds. The report shows 1,000 − 50 = $950 in net item sales and explicitly excludes taxes and shipping. It does not call $950 profit, ad ROI, or a customer result. Completion is seeing the source rows and calculation. A real case-study CTA would require a real customer, method, period, and outcome instead.
6. Make comparison criteria visible before declaring a winner
Customer question: “Which lamp fits a desk without an electrical outlet?”
Mismatch: Find the best lamp for every desk leads to two cards that emphasize the more expensive model but omit the trade-offs.
Better ad: “Compare Cord and Go by power source, light output, and price. See the trade-offs before choosing a lamp.”
CTA: Compare the two lamps
The destination compares the same fields: Cord costs $48, uses mains power with a 1.8-meter cable, and is rated at 500 lumens; Go costs $68, charges by USB-C, and is rated at 300 lumens. Go can be used at a desk without an outlet while charged, but its battery runtime is not specified. That makes Go a candidate for the stated power constraint, not a proven all-day choice. The comparison is complete when it exposes both the relevant difference and the missing information.
7. Call a sample preview a sample preview
Customer question: “Can I see the workflow before creating an account or booking a call?”
Mismatch: Try it with your team’s data opens a fixed walkthrough that cannot import data or perform a real action.
Better ad: “Preview a support workflow with three sample tickets. No account needed. Explore the example without importing your team’s data.”
CTA: Open the sample demo
The destination lets the visitor select sample ticket A002, add an “Account access” tag, generate a sample password-reset response, and finish with a notice that no customer data was used and no message was sent. This is a preview, not a trial. A trial would need actual product access; a personalized demo would need a different promise and perhaps a booking step. The companion CTA Continuity Checker includes this worked fictional flow as an interactive example.
8. State what “no credit card” does—and does not—remove
Customer question: “Can I use the product before paying, and what happens when the trial ends?”
Mismatch: Try it now—no signup leads to required account creation and email verification, or worse, a waitlist.
Better ad: “Use the product for 14 days after creating and verifying your account. No credit card is required, and the trial does not charge you automatically.”
CTA: Start the 14-day trial
If activation occurs at 09:00 UTC on October 1, 2026, a 14-day trial ends at 09:00 UTC on October 15—a span of 14 × 24 = 336 hours. Before verification, the status should say the trial has not started. At expiry, it should say no charge was made and offer a separate $49-per-month checkout. “No card” is not the same promise as “no account,” and a submitted form is not the same event as trial_active.
9. Disclose a resource gate instead of hiding it
Customer question: “Can I get a usable checklist, and what information do I have to provide?”
Mismatch: Download instantly—no signup leads to a required email form; after submission, the visitor sees only “Thanks.”
Better ad: “Get a free five-point CTA checklist. Enter your email to access it after the form; newsletter updates are optional.”
CTA: Get the free checklist
The destination asks only for email, leaves the newsletter opt-in unchecked, and reveals the resource on the same page after submission. The checklist names the job, states the conditions, follows the real path, checks an unavailable state, and defines completion separately from the click. Its success message distinguishes “resource available now,” “email copy requested,” and “email delivered.” A gate is not automatically good or bad; the error is promising ungated access and delivering a gate—or collecting the lead without delivering the resource.
10. Do not call a request a booked appointment
Customer question: “Can I choose a time to get help with setup?”
Mismatch: Your appointment is booked appears after a general contact form with no selected time.
Better ad: “Choose an available time for a free 20-minute setup call. You’ll receive a confirmation after selecting your slot and entering your name and work email.”
CTA: Choose a setup-call time
The destination shows the time zone, available slots, the 20-minute duration, required name and work email, and optional company field. A fictional selection of September 15, 2026, 10:00–10:20 at UTC−04:00 corresponds to 14:00–14:20 UTC. Confirmation appears only after that slot is reserved. If the page can collect only interest, use Request a setup call and say the time is not yet confirmed. Empty availability is another valid state; it must not masquerade as a booking.
11. Let a ready buyer shop
Customer question: “I’ve chosen the gray organizer. Where do I review the cost and order it?”
Mismatch: A product-specific ad sends the buyer to a style quiz, a generic catalog, or a page that adds undisclosed shipping.
Better ad: “The gray desk organizer is $24, with $5 shipping to the contiguous US. Review the product and costs before checkout. Taxes are extra.”
CTA: Shop the gray organizer
The product page opens with the gray SKU selected, shows the price, shipping and return terms, and lets the buyer add that item to the cart. The before-tax review is $24 + $5 = $29; it should not be called the final total before tax is known. This is a case where a direct purchase CTA is appropriate. Sending a shopper who has already chosen the item through a guide or demo would add a new task rather than answer the stated question.
12. Preserve the known variant through a reorder
Customer question: “Can I buy the same whole beans again without accidentally subscribing?”
Mismatch: Reorder your whole beans opens a general coffee catalog, changes the grind, or defaults to a monthly subscription.
Better ad: “Reorder your 12 oz whole beans. Review today’s price and quantity before paying. This reorder is a one-time purchase, not a subscription.”
CTA: Review your one-time reorder
After any required login, the destination returns to the 12-ounce whole-bean SKU, shows two bags at $18 each, adds $5 shipping, and labels the purchase as one-time: 2 × $18 + $5 = $41 before tax. If the previous SKU is unavailable, the page says so instead of silently substituting a grind or subscription. Login can be a legitimate condition when personal order history is required; continuity means preserving the intended reorder after that condition is satisfied.
What real ad-to-destination chains reveal
The fictional examples make the mechanics visible, but real chains add two complications: links change over time, and a broad CTA may be fulfilled at one level of the journey without completing a later transaction.
| Observed chain | Promise in the dated ad | What was available on September 9, 2026 | What it teaches |
|---|---|---|---|
| Contentful sponsor in TLDR Marketing, May 5, 2026 | “Get your (ungated) download” | The click opened a Contentful report page, which linked to an accessible 28-page report PDF without a form in the inspected path. | Verify the artifact itself, not merely the landing-page headline. This observation says nothing about ad performance or the report’s research quality. |
| LaunchDarkly sponsor in TLDR Tech, December 15, 2022 | Complete an interactive feature-flag demo; “Try it now for free” | The archived /test-drive/ link now redirects to a request-a-demo form requiring work email, name, company, and job title; the form was not submitted. |
Audit old campaign links. The current redirect no longer provides the self-serve step described in the 2022 ad, but it does not prove the destination was mismatched in 2022. |
| Quince sponsor in The Cooking Foodie, June 28, 2025 | “SHOP NOW” for home products | The archived SHOP NOW affiliate link could not be completed in this check. The clean current Quince decor route opened a US product grid with filters, prices, and product links. No cart or checkout was completed. | “Shop” can legitimately land on a category. That fulfills browsing, not purchase completion, and the exact historical tracking chain remains only partially observed. |
An accessible chain is not evidence that an ad was effective. It only establishes whether the promised step was present in the path we could inspect, on the date we inspected it.
Platform CTA controls are not interchangeable
The sentence in the ad, the platform’s native CTA control, and the button on the destination are separate surfaces. Their rules also vary by format and objective, so a universal list of “available buttons” is misleading.
| Platform surface | Current documented control | Continuity implication |
|---|---|---|
| Google responsive search ads | Google says an RSA can contain up to 15 headlines and four descriptions, and that assets can appear in different combinations and orders. Its preview does not show every combination. See About responsive search ads. | Write each eligible asset so it remains truthful in combination, then verify the final URL and any asset-level destination behavior. One preview is not a complete QA pass. |
| TikTok Ads Manager | TikTok documents Dynamic CTA, AI-generated CTA based on the landing page or target app, and fixed Text CTA. Availability depends on objective; with CTA enhancement enabled, the system may generate additional CTA text during the campaign. See Dynamic & Recommended CTA, updated November 2025. | Audit generated and dynamically selected labels as well as the copy you typed. Do not claim every account or objective exposes the same control. |
| LinkedIn single-image and lead-generation ads | LinkedIn’s single-image specifications say the objective can produce either a destination URL or Lead Gen Form and that the CTA is selected from available options. With the lead-generation objective, clicking the CTA opens a native form, according to LinkedIn’s lead-generation instructions. | The fulfillment may happen inside the platform rather than on a website. Inspect the native form fields, promise, success state, and downstream delivery—not only the landing page. |
The operational rule is not “always use the same CTA everywhere.” It is “identify which surface controls the promise, then inspect every eligible path that can be served.”
A CTA test must say what actually changed
A good-looking button is not causal evidence. Even a documented test needs its offer, destination, audience, allocation, optimization, and outcome defined.
AdEspresso’s 2018 vendor test is useful precisely because its scope is visible. It copied one Facebook lead-generation ad into four separate ad sets and changed the native CTA among no button, Learn More, Download, and Sign Up. Each cell spent $250 over 11 days on US mobile News Feed traffic from a 1% lead-based lookalike audience, optimized to a thank-you-page lead pixel. The article reported 20, 36, 49, and 26 leads respectively. See What’s the Best CTA for Your Facebook Ads?.
That result does not establish Download as a universal winner. It concerns one ebook offer, one historical platform environment, separate ad sets, and a lead event with no reported downstream quality. It also contains an internal arithmetic problem: $250 divided by 36 Learn More leads is about $6.94, not the $9.94 CPL written in the article. The lead counts are more defensible than that printed CPL line.
Your own test should name the result that matters after the click. This synthetic example shows why:
| Synthetic variant | Impressions | CTA clicks | Completed resource opens | CTA CTR | Opens per click | Opens per impression |
|---|---|---|---|---|---|---|
| A | 10,000 | 300 | 90 | 3.0% | 30% | 0.9% |
| B | 10,000 | 200 | 120 | 2.0% | 60% | 1.2% |
Variant A gets more clicks; Variant B produces more completed opens. Without spend, sales, lead quality, random assignment, and a defined attribution window, neither row is a commercial winner. The example only proves an arithmetic point: optimizing the click can rank variants differently from optimizing the promised task.
Run a five-minute continuity audit before launch
- Name the customer question and its evidence status. Cite the research source or mark it as a hypothesis. Do not upgrade “plausible” to “biggest” without comparative evidence.
- Write the exact promise. Decide whether the next step is to inspect a price, calculate an estimate, check fit, view proof, compare, preview, start a trial, access a resource, book, buy, or reorder.
- Disclose the conditions that change the decision. Include required accounts, email gates, cards, region, price basis, timing, data requirements, and whether a quote or appointment is confirmed.
- Follow the real served path. Test the ad URL, redirects, native form, landing page, and final artifact or product state. For dynamic formats, inspect more than one preview.
- Try a negative or unavailable case. Use an unsupported model, out-of-area ZIP, missing slot, unknown input, expired trial, or out-of-stock variant. A truthful “not available” can fulfill the check better than an invented positive answer.
- Define completion separately from the click. Examples include
price_card_viewed,estimate_generated,resource_opened,trial_active,slot_confirmed, andpaid_order. Track later quality or revenue separately when it matters.
When the audit fails, there are only two honest repairs: change the ad to describe the step the destination really offers, or change the destination so it fulfills the promise. Swapping Learn More for Get Started while leaving the mismatch intact is not a strategy.
The best CTA is not always the lowest-commitment option. A shopper who has selected a SKU may need Shop the gray organizer. An existing customer may need Review your one-time reorder. Someone who needs a price may need a calculator rather than a sales call. Match the commitment to the job, make the conditions visible, and judge the whole path by whether the promised result appears.
Sources
- “Get Started” Stops Users — Nielsen Norman Group, August 20, 2017.
- Learning about users and their needs — GOV.UK Service Manual, published April 4, 2016; updated March 23, 2017.
- TLDR Marketing: May 5, 2026 — dated Contentful sponsor placement.
- When machines make marketers more human — Contentful resource page.
- When machines make marketers more human — PDF — Contentful and Atlantic Insights; resource access inspected, report findings not evaluated here.
- TLDR Tech: December 15, 2022 — dated LaunchDarkly sponsor placement.
- Request a Demo — LaunchDarkly; current redirect destination inspected September 9, 2026.
- New Recipe + Cozy Favorite + Something Cool Inside — The Cooking Foodie, June 28, 2025; Quince sponsor placement and affiliate disclosure.
- All Home Decor — Quince; current US category route inspected September 9, 2026.
- About responsive search ads — Google Ads Help; checked September 9, 2026.
- Dynamic & Recommended CTA — TikTok for Business, updated November 2025; checked September 9, 2026.
- Single image ads advertising specifications — LinkedIn Marketing Solutions Help; checked September 9, 2026.
- Create LinkedIn lead generation ad sets — LinkedIn Marketing Solutions Help; checked September 9, 2026.
- What’s the Best CTA for Your Facebook Ads? — AdEspresso, November 1, 2018.
