AD FIELD NOTESAI advertising · the work behind the output

Research library / September 2026 notebook

Choose a tool by its job

A dated comparison of creative tools, terms, privacy and production trade-offs.

Research snapshot dated 14 September 2026. Prices, product access and policy details can change. Check the linked primary sources for a new brief or release decision.

This section is a representative comparison, not a promise that one tool is “best.” Price and product details were checked against official pages where available on September 14, 2026. Exact commercial rights depend on plan, jurisdiction, inputs, model, output, and client contract.

Current price and rights snapshot

These are practical buying signals, not a substitute for the plan page or a client contract. Credit allowances, model access, regional availability, and commercial terms change quickly.

ToolPoint-in-time price signalCommercial / privacy signalDecision note
ChatGPT BusinessStandard listed at $20/user/month billed annually or $25 monthly; Premium listed at $100 annually or $125 monthly; self-serve seats and API are separateOpenAI says business inputs/outputs are not used for training by default; enterprise controls varyGood generalist for briefs, research synthesis, image work, and structured review; preserve citations and keep sensitive inputs in the right workspace
ClaudePro and team/enterprise tiers; regional and usage terms varyAnthropic says commercial customers control data and commercial data is not used to train generative modelsStrong long-form strategy and document work; do not confuse careful prose with verified facts
MidjourneyBasic $10/month, Standard $30, Pro $60, Mega $120; annual billing is discountedUsers own images/videos to the fullest extent possible under its terms; businesses above $1M gross revenue need Pro or Mega; public/remix defaults matterExcellent visual exploration; weaker fit for confidential or tightly deterministic production unless the plan and workflow are controlled
Adobe FireflyCurrent page lists Standard $9.99, Pro $19.99, Pro Plus $49.99, Premium $199.99/month, with credit allowances that differ by tierAdobe positions its own Firefly models as trained on licensed/public-domain material and supports Content Credentials; partner models require separate reviewStrong choice when Photoshop, Premiere, After Effects, brand assets, and review already live in Adobe
RunwayFree tier; Standard $15 monthly/$12 annual, Pro $35/$28, Max $95/$76; credit and model limits differRunway says users own and may use generated content commercially, subject to inputs, terms, and other rightsStrong video ideation and production workflow; budget credits and consistency tests before promising volume
Veo / Vertex AIAPI pricing is model- and modality-dependent; a current example is per-second video billing, so model the shot list before committingCommercial use, data handling, region, and access depend on Google product and contractUseful for high-quality motion, native audio, and controlled references; treat model-page claims as vendor claims
SoraNo current production price: OpenAI’s page says the Sora product was no longer available as of April 26, 2026Historical terms and provenance controls should not be treated as current availabilityKeep Sora in the history of AI advertising; do not build a 2026 dependency on it
ElevenLabsFree; Starter $6, Creator $22, Pro $99, Scale $299, Business $990 listed at the current snapshotPaid plans include commercial use; free output is non-commercial and attribution-based; voice-clone authorization remains separateUseful for drafts and localization; log voice consent, language, territory, and usage scope
HeyGenCredit-based plans; the product page and API page are the source of truth for current seats, minutes, and translation creditsDigital-twin and voice use depends on consent, plan, and agreementPractical for multilingual spokesperson workflows, but stock-avatar and paid-ad restrictions must be checked per use
Synthesia licensingFree/basic through enterprise; enterprise pricing is customStandard stock/synthetic avatars cannot be used in paid TV or paid social advertising; custom avatars require informed consent and agreementA crucial “not automatically ad-ready” example; choose the avatar license before writing the media plan

The biggest buying mistake is comparing generation price alone. A defensible total cost includes source-asset preparation, prompt or storyboard labor, cleanup, rights and consent, native review, legal, localization, version QA, storage, provenance, and the media experiment. A cheaper model can be more expensive if it creates more rejected work or weaker auditability.

Strategy, research, copy, and planning

Tool / platformPrimary useBest fitCurrent snapshotRights / privacy watchoutEditorial view
ChatGPT Business / EnterpriseResearch, writing, analysis, image, structured workflowsAgencies, marketing teams, foundersBusiness standard seat listed at $20/user/month annually or $25 monthly; Premium at $100 annual or $125 monthly; minimum two seats on self-serve BusinessOpenAI says business inputs/outputs are not used for training by default; consumer settings differ; API billed separatelyStrong generalist for briefs, synthesis, and reviewable workflows; keep source citations and structured outputs
ClaudeLong-form analysis, writing, research, coding, document reviewStrategy teams and editorial workClaude Pro listed at $20/month US; Max tiers $100 and $200; commercial Team/Enterprise data is not used to train generative models per AnthropicCommercial customer controls data as controller; Anthropic acts as processor for commercial productsGood for careful writing and long context; still requires source verification and policy review
Gemini / Google WorkspaceResearch, copy, documents, Sheets, video ecosystemGoogle-native teamsGoogle AI Plus, Pro, and Ultra personal plans; Workspace AI is attached to Workspace plans/add-ons; regional features and prices varyGoogle says Workspace protections apply and content is not used to train outside the domain without permissionStrong when campaign context already lives in Drive/Docs/Sheets; permissions need mapping
JasperBrand-governed marketing copy and campaign operationsMarketing teams with a brand systemSubscription pricing and feature access change; verify current plan and usage limitsReview data processing, model providers, and ownership termsUseful when the content workflow is the product; not a substitute for strategy
Copy.aiGTM and marketing workflow automationRevenue and content teamsSubscription/usage model; verify current pricing and API termsScope data access, connectors, and generated claimsGood for repeatable workflows; avoid “autopilot” publishing without approval
HubSpot BreezeCRM, content, service, and marketing assistanceSMB and mid-market teams using HubSpotIncluded/plan-dependent plus credits and HubSpot subscriptionCRM exposure and customer-data governance matter more than model nameStrong integration value if HubSpot is already the system of record
Salesforce Einstein / AgentforceCRM actions, service, sales, marketing, and agentsEnterprise CRM teamsContract, credits, and seats; public entry price varies by productData Cloud, permissions, prompt/output handling, and action scope require contract reviewPowerful for governed workflow orchestration; expensive to configure well

OpenAI’s current business privacy page says it does not use business data for training by default and offers retention, data residency, and admin controls. Anthropic makes a similar commercial-data distinction, explicitly separating consumer products from commercial controller/processor terms.1516 The lesson is to choose the plan and data boundary, not just the model brand.

Image, design, video, and editing

Tool / platformPrimary useCreative controlCommercial / privacy snapshotCost signalMain weakness
Adobe Firefly + Photoshop / Premiere / After EffectsImage generation, editing, vector, video, audio, resizing, content supply chainHigh inside Adobe workflows; supports references, style kits, editing, and brand controlsAdobe states Firefly models are trained on licensed/public-domain content, outputs are intended for commercial use, and Content Credentials are attached; partner models require separate reviewFirefly Premium listed at US$199.99/month per license with 50,000 credits on the current page; lower plans exist“Commercially safe” does not clear your inputs, claims, likenesses, or partner-model output
MidjourneyImage and video generationStrong aesthetic exploration, weaker deterministic production controlUsers own assets to the fullest extent possible under terms; companies over $1M revenue need Pro/Mega for commercial use; public work is viewable/remixable by default unless Stealth appliesSubscription tiers; check current pricingPublic-by-default workflow, reference similarity, and limited enterprise governance
RunwayVideo, image, audio, editing, API generationStrong creative iteration and multimodal workflow; API supports per-second billingRunway says users retain ownership and commercial use rights in their generated content; third-party rights and inputs remain your responsibilityAPI credits cost $0.01 each; Gen-4 Turbo listed at 5 credits/sec; web plans include monthly credit allotmentsCredits expire on many plans; model and quality tradeoffs can make costs unpredictable
SoraText/image-to-videoHistorically strong demonstration of generative videoOpenAI’s current Sora page says the Sora product was no longer available as of April 26, 2026; do not assume access or build a current production dependencyAvailability caveat is the important cost signalProduct discontinuity and evolving terms
Veo / Google FlowVideo, image-to-video, native audio, scene extensionGoogle describes reference images, camera controls, character consistency, and audioVerify access, commercial terms, and region through Google AI/Cloud product terms; benchmark claims on the model page are vendor studiesPlan/API dependent; no single universal priceModel availability, policy, and output terms vary by surface
Kling / Pika / LumaGenerative video and image-to-videoUseful for exploration and short-form effectsVerify current plan, training/data, watermark, and commercial terms before client workSubscription/credit models change quicklyPublic benchmark quality is not the same as production reliability
CanvaDesign, templates, brand kits, AI copy/image/videoEasy, fast, accessible; less granular than a specialist pipelineCanva’s published Magic Studio terms require review and clear disclosure of AI output; output is used at user’s risk; current terms should be checked because the cited page is an archiveFree and paid plans; AI feature limits varyTemplate gravity can flatten brand distinctiveness
DaVinci ResolveEdit, color, audio, VFX with optional AI featuresHigh human editorial and finishing controlRights depend primarily on source assets and plug-ins; verify Studio feature setOne-time/plan pricing varies by version and regionAI is not a shortcut for an editorial point of view
DescriptTranscript-based editing, captions, voice, clipsHigh speed for talking-head and social workflowsCurrent Pro page lists $24/editor/month annual-equivalent or $30 monthly, 4K export, 30 transcription hours, and unlimited Overdub useSubscription plus transcription/model limitsLess suited to complex visual finishing or high-end sound

Adobe’s commercial-safety position is a vendor claim with useful documentation, not a universal warranty. Adobe explicitly says partner-model outputs need separate review and that users remain responsible for lawful use.17 Midjourney explicitly places ownership subject to third-party rights and requires large businesses to use Pro or Mega; its public/remix defaults matter for confidential campaign work.18 Runway’s own usage-rights page is comparatively clear about commercial use, but it also does not make third-party rights disappear.19

Voice, avatars, sound, and localization

Tool / platformPrimary useBest fitCurrent rights / cost snapshotGuardrail
ElevenLabsTTS, voice cloning, dubbing, sound effects, musicVoiceover, localization, audio prototypingStarter listed at $6/month and includes commercial license; Creator $22, Pro $99, with higher team tiers; free plan is non-commercial with attributionConsent and scope for every cloned voice; paid plan rights do not authorize a person’s likeness
HeyGenAvatars, digital twins, translation, lip-sync, interactive videoMultilingual spokesperson and internal/external videoCreator $29/month, Pro $49, Business $149 plus seats; free output is non-commercial; custom voice/likeness requires ownership or consentStock-avatar licenses may differ from custom avatars; paid advertising use must be checked in plan terms
SynthesiaStock and custom avatars, translation, training/product videoEnterprise explainers, learning, controlled spokespeopleSynthesia says stock/synthetic avatar videos cannot be used in paid TV or paid social advertising under the standard license; custom avatars can be used under the person’s agreement with consentDo not assume a stock avatar is an advertising endorser
Descript OverdubVoice clone and transcript editingFast social/podcast/video editingIncluded in plan; current Pro page lists unlimited OverdubUse only a voice you own or have explicit permission to clone
Spotify AI ad creationAI voice, tone, music versions for audio adsPerformance audio testingPlatform feature/case-study model; media and creative access depend on Spotify AdsVerify talent, music, and platform disclosure requirements
Adobe / Premiere / Firefly audioCleanup, speech, sound effects, music, translationAdobe-centered post and content supply chainCredit and feature dependent; Firefly-specific commercial claims do not automatically cover partner modelsKeep the original track, license, and model record

ElevenLabs states paid plans include commercial rights for generated audio while the free plan is non-commercial and attribution-based; the commercial right still sits inside the service terms and does not replace voice consent.20 HeyGen’s current terms distinguish free-plan non-commercial output from paid plans and its product guidance says voice cloning begins with consent. Synthesia’s licensing page is an important counterexample: stock and synthetic avatars are restricted from paid advertising, while custom avatars require free, informed consent and a separate agreement.122122

Media platforms, personalization, and optimization

PlatformWhat AI does for advertisersCurrent evidence / featureWhat to verify
Google Ads Performance Max / AI MaxGenerates headlines, descriptions, images, videos, extensions, and optimizes delivery across Google inventoryGoogle says advertisers can review generated assets, generate bespoke images, and use automatic video if they do not upload one; Google also says AI-generated images carry SynthID/metadataEligibility, sensitive-category restrictions, generated-asset labels, landing-page expansion, and actual delivery
Meta Advantage+Automated audience, placement, creative enhancement, image/text generation, and deliveryMeta says “About this ad” includes AI info labels for ads created or significantly edited with its tools and is expanding transparency for third-party AI toolsCreative settings, labels, audience expansion, brand-safe placements, and incrementality
TikTok SymphonyIdeation, trend research, video generation, translation/dubbing, refresh, resizing, optimizationTikTok says Symphony tools are integrated with Ads Manager and Creative Center; some 2025 case figures are vendor-reportedCreative GenAI terms, labels, voice/likeness, content safety, and spend controls
Amazon AdsProduct-grounded lifestyle image generation and creative variationAmazon says Image Generator uses product images/descriptions/reviews; Amazon reports increased submissions and GMS in its own case materialProduct accuracy, rights to inputs, marketplace policy, and whether vendor lift generalizes
Spotify Ads ManagerAI-generated audio variations and testingSpotify case study reports better traffic/cost for SHI InternationalVoice/music rights, audience split, and full test design

Google explicitly states that generative asset creation can be optional and reviewable, but eligibility excludes some sensitive categories and the generated asset feature may be unavailable in some languages or regions.1 Meta describes labeling for its own generative creative and expanding transparency for third-party AI tools.23 TikTok’s Symphony materials describe ideation, translation, avatars, and optimization but also state that past performance does not guarantee future performance.2 Amazon’s image-generation result figures are Amazon-reported and should be labeled as such.4

Follow the evidence

This chapter comes from the September research notebook. Linked sources and qualifications remain attached to the claims.

Open all sources and footnotes →

The weekly field note

One useful note for the next brief.

Campaigns worth studying, the work behind them, and questions to bring into your next creative review. Delivered weekly.

How we handle your email