---
url: "https://auraplusplus.com/blog/generative-ai-for-advertising"
markdown: "https://auraplusplus.com/blog/generative-ai-for-advertising.md"
category: "generative ai for advertising"
type: "outrank"
published: "2026-09-06"
word_count: "n/a"
---

# Generative AI for Advertising: A Practical Guide

> Learn how generative AI for advertising is reshaping creative, targeting, and measurement. Practical workflows, use cases, risks, and ROI tips.

## Article

You launch a batch of AI-generated ads on Monday. By Wednesday, impressions are climbing, the platform dashboard is full of activity, and your team has more creative than it could have produced in a month. Then someone asks the question nobody prepared to answer: **Which ad generated the qualified signups?**

The answer is usually buried under clicks, reach, video views, audience breakdowns, and platform-reported conversions. The creative pipeline moved faster, but the growth system didn't become more intelligent. That distinction defines the opportunity in **generative AI for advertising**. AI can produce more possibilities, but only disciplined measurement tells you which possibilities deserve more budget.

Generative AI has already moved into routine advertising work. The practical challenge now isn't adoption. It's connecting AI-assisted production to **incremental revenue, efficient customer acquisition, stronger creative-market fit, and durable visibility beyond a paid campaign**.

## Table of Contents

- [The Moment Every Founder Hits With Their First AI Ad Campaign](#the-moment-every-founder-hits-with-their-first-ai-ad-campaign)
- [What Generative AI in Advertising Actually Means](#what-generative-ai-in-advertising-actually-means) [The four operating roles](#the-four-operating-roles)
- [What it isn't](#what-it-isnt)

- [Where Generative AI Already Changes Ad Campaigns](#where-generative-ai-already-changes-ad-campaigns)

- [Why Faster Production Is Not the Same as Better Results](#why-faster-production-is-not-the-same-as-better-results)
- [Three places where AI can create real lift](#three-places-where-ai-can-create-real-lift)

- [How Maxfusion AI Can Help](#how-maxfusion-ai-can-help)

- [Real Benefits and Real Risks Side by Side](#real-benefits-and-real-risks-side-by-side)
- [The controls that make the trade-off work](#the-controls-that-make-the-trade-off-work)

- [A Practical Implementation Workflow for Startups and Marketers](#a-practical-implementation-workflow-for-startups-and-marketers)
- [Build the operating foundation](#build-the-operating-foundation)
- [Design tests that can teach you something](#design-tests-that-can-teach-you-something)
- [Close the feedback loop](#close-the-feedback-loop)

- [Optimization and Measurement Tactics That Separate Winners](#optimization-and-measurement-tactics-that-separate-winners)
- [Test one meaningful idea at a time](#test-one-meaningful-idea-at-a-time)
- [Improve the signal before improving the model](#improve-the-signal-before-improving-the-model)

- [Your First 30 Days and What to Watch Next](#your-first-30-days-and-what-to-watch-next)
- [Week one, audit and goal setting](#week-one-audit-and-goal-setting)
- [Week two, prompt library and creative sprint](#week-two-prompt-library-and-creative-sprint)
- [Week three, structured testing](#week-three-structured-testing)
- [Week four, review and scale](#week-four-review-and-scale)

## The Moment Every Founder Hits With Their First AI Ad Campaign

A founder at an early-stage software company opens the campaign dashboard after a week of testing. The team has generated product demos, founder-style videos, testimonial-style scripts, static graphics, and several variations of the same offer. Every asset has a neat filename and a different hook.

The top-line numbers look healthy. Reach is expanding. The ad platform reports engagement. A few ads appear to be winning on click-through rate. But the signup report tells a different story. Paid traffic is arriving, yet qualified accounts haven't increased enough to justify the spend. The founder starts asking whether the winning ad is persuasive or attracting low-intent curiosity.

That moment creates a dangerous temptation. The team can respond by generating another hundred variants, changing thumbnails, rewriting headlines, and adding more audience segments. Production feels like progress because production is visible. Revenue attribution is slower, messier, and less flattering.

> Practical rule: Never treat the number of assets produced as evidence that the campaign improved.

The gap isn't theoretical. The [IAB Europe and Microsoft 2024 report on AI in digital advertising](https://iabeurope.eu/wp-content/uploads/IAB-Europe-and-Microsoft-AI-in-Digital-Advertising-Report-2024.pdf) found that **91% of respondents had used, were using, or had experimented with generative AI**, while **38% said it was becoming embedded in daily work**. That adoption tells you the tools have entered normal operations. It doesn't prove that the resulting campaigns create business value.

The founder's real problem is therefore not “How can we make more ads?” It's “How can we identify the creative, audience, offer, and post-click experience that caused a meaningful business outcome?”

This guide takes that position seriously. AI is useful when it expands the number of strategically different ideas a team can test, improves relevance, and creates a traceable feedback loop. It becomes expensive theatre when it produces interchangeable assets that nobody can evaluate cleanly.

## What Generative AI in Advertising Actually Means

**Generative AI in advertising** refers to models that create or modify campaign inputs and outputs from prompts, brand information, audience data, product feeds, and performance feedback. Those outputs can include copy, images, video, voice, audience insights, creative concepts, and media recommendations.

A useful analogy is an in-house creative department paired with a media buyer. The creative team turns a product brief into headlines, storyboards, images, and videos. The media buyer chooses placements, bids, and audience signals. A generative AI system can assist with both roles, but it still needs direction, constraints, and a reliable definition of success.

![A diagram explaining the four key roles of generative AI in advertising: creative, audience, media, and learning.](https://cdnimg.co/b866be35-93f2-4b64-91bc-8c253a419ab8/262d4455-29e3-43c8-b66d-1e619889c7d2/generative-ai-for-advertising-ai-diagram.jpg)

### The four operating roles

1. Creative production turns a brief into variations. A team might provide product positioning, approved claims, visual references, and a target segment. The model can then produce headline families, image directions, video scripts, or finished assets.
2. Audience signals help describe who should see an ad and why. AI can organize customer research, identify recurring pains in interviews, cluster language from support conversations, and translate those signals into audience hypotheses. It shouldn't be treated as a substitute for consent, data governance, or human judgment.
3. Media decisions involve placement, bidding, pacing, and allocation. Many ad platforms already use machine learning to make these decisions. Generative AI adds a conversational layer and can help teams interpret options, prepare inputs, or generate testing plans, but platform automation still requires independent business measurement.
4. Continuous learning connects results to future work. A strong workflow records which message, visual, audience, placement, and landing-page experience produced qualified outcomes. The next prompt should use those learnings instead of merely asking for more novelty.

### What it isn't

Predictive AI estimates what may happen, such as the likelihood of conversion. Analytics describes what happened. Rule-based automation follows predefined instructions, such as pausing an ad after a threshold. Generative AI produces new material or recommendations.

The distinctions matter because each system can optimize a different layer. A model can generate a polished video without knowing whether the offer is credible. A predictive model can score leads without explaining why the creative failed. Your campaign needs a clear handoff between generation, prediction, activation, and measurement.

## Where Generative AI Already Changes Ad Campaigns

The strongest applications remove bottlenecks that previously forced small teams to choose between speed and variety. They don't eliminate strategy. They make strategic testing less expensive and less dependent on production schedules.

For a product launch, a team can turn one positioning brief into multiple hook families. One family may lead with the problem, another with a workflow demonstration, and another with a category comparison. The value comes from preserving a distinct hypothesis in each group, not from changing adjectives across near-identical ads.

Paid social benefits from persona-level adaptation when the underlying promise stays accurate. A productivity product might express the same benefit through different situations for a solo developer, an agency operator, or a finance lead. The marketer supplies the approved claims and audience insight. AI supplies the language variations, opening scenes, and format adaptations.

Retail teams can use generated lifestyle imagery to explore contexts that are difficult to stage repeatedly. The safe version grounds the output in real product photography, approved visual references, and a clear review process. The unsafe version invents product details, packaging, or usage situations that the buyer won't receive.

Video remixing is another practical use. Existing footage can be recut into shorter openings, alternate captions, different aspect ratios, or new voiceover structures. The tool saves editing time, but the team still needs to confirm that every claim, subtitle, transition, and product demonstration remains accurate.

Dynamic product advertising can combine catalog feeds with generated copy and layout variations. AI-assisted keyword and audience discovery can also turn customer language into testable search themes and targeting hypotheses. Those outputs belong in a testing plan, not directly in a campaign without review.

Use Case
Ad Format
Workflow Stage

Hook and headline generation
Search, social, display
Concept development

Persona-level message adaptation
Paid social, email retargeting
Creative production

Grounded lifestyle imagery
Retail, display, social
Visual production

Video remixing
Short-form video, streaming video
Editing and versioning

Product-feed assembly
Dynamic product ads
Activation

Keyword and audience discovery
Search and social
Research and planning

For teams building video variations from an existing library, an [AI video ad generator for campaign production](https://auraplusplus.com/projects/ai-video-ad-generator) can fit the versioning stage. It won't determine whether the offer is strong or whether the landing page converts. Its role is narrower, and that can be useful when the production bottleneck is clearly identified.

The evidence points to broad adoption across formats. In [IAB's 2025 Digital Video Ad Spend and Strategy coverage](https://www.iab.com/insights/the-ai-gap-widens/), **86% of buyers said they were using or planning to use generative AI to build video ad creative**. The same coverage reported that **22% of video ad creative had already been built or enhanced with generative AI in 2024**, with a projected **39% by 2026**. It also described substantial use in social and display advertising, with lower use in television and audio.

The use case is not “AI made an ad.” It's a workflow that lets a team test meaningfully different creative ideas, learn which ones attract valuable customers, and reuse the winning insight across formats.

## Why Faster Production Is Not the Same as Better Results

A startup can produce two hundred ad variants in a week and still fail to improve conversion. If every variant uses the same weak offer, vague positioning, undifferentiated audience, and confusing landing page, the team has multiplied the presentation of the problem.

Production metrics answer operational questions. How many assets did the team render? How quickly did it deliver them? How many formats and sizes are available? Business metrics answer different questions. Did qualified acquisition become more efficient? Did the campaign create incremental demand? Did the brand become more memorable or trusted?

Those metrics can move independently. A high click-through rate may reflect curiosity, a misleading visual, or an audience that isn't ready to buy. A low-cost impression may deliver little commercial value. A polished video may fail because the first seconds don't communicate the product's relevance.

![An infographic illustrating the gap between fast generative AI asset production and effective business results.](https://cdnimg.co/b866be35-93f2-4b64-91bc-8c253a419ab8/e3ba24d5-11e0-4cba-8a5b-ea9d4943ead3/generative-ai-for-advertising-business-results.jpg)

### Three places where AI can create real lift

**Creative-market fit** comes first. The asset must express a problem, desire, or outcome that matters to the intended audience. AI helps by making more concepts feasible, but humans must decide which customer truth each concept represents.

**Auction-level relevance** comes next. The right message must reach a user in a context where it makes sense. Audience signals, placement, bid strategy, and frequency influence whether a strong creative gets a fair opportunity.

**Post-click experience** closes the loop. If the ad promises a specific outcome and the landing page offers a generic explanation, the campaign loses momentum. AI-generated copy can't repair a broken signup path or an offer that lacks proof.

The [Columbia research on AI-generated display ads](http://www.columbia.edu/~on2110/Papers/AI_in_disguise.pdf) illustrates why creative quality and perception matter. In the reported quasi-experimental setting, AI-generated ads recorded an average **CTR of 0.76% across 885,434 clicks from 117,002,984 impressions**, compared with **0.65% for human-made ads**. Ads with obvious AI cues underperformed, which indicates that the audience response depended on execution and perception, not on the presence of automation.

The practical lesson is uncomfortable but useful. More output expands the search space. It doesn't tell you where the answer is. Your measurement design has to connect each creative hypothesis to a business outcome, while your review process protects the brand from synthetic-looking or unsupported claims.

## How Maxfusion AI Can Help

Maxfusion AI is designed for performance teams that want research, ideation, generation, and assembly in one environment. Its MaxFlows canvas connects those stages visually, while MCP connectors let teams operate the workflow through agent interfaces such as Claude, ChatGPT, Cursor, or Hermes.

That consolidation solves a real operational problem. A fragmented stack forces marketers to move research into one tool, prompts into another, image generation into a third, video production into a fourth, and editing into a fifth. Maxfusion AI brings together competitor discovery through the Meta Ad Library, TikTok trend analysis, structured hooks and scripts, image and video models, compositing, captions, and export.

![Screenshot from https://maxfusion.ai](https://cdnimg.co/b866be35-93f2-4b64-91bc-8c253a419ab8/screenshots/c27d72e8-e123-437f-abb0-7be3581e3ab6/generative-ai-for-advertising-maxfusion-homepage.jpg)

Its model aggregation is useful when a team needs to compare creative styles rather than commit to one generation engine. The platform lists multiple image and video models, plus RIZZ for audio-guided UGC-style videos, an Actor Library with **300+ AI actors**, consent-based actor creation, and support for **35+ languages**. CopyLab, Google Drive integration, a public API, and editing tools such as lip sync, voice cloning, background removal, trimming, and stitching support production at scale.

The trade-off is straightforward. A unified platform reduces context switching and can help teams batch work, but it doesn't replace creative direction, legal review, consent management, or incrementality testing. The credit-based plans and model choices also mean a team should estimate its actual production pattern before committing.

[Maxfusion AI](https://maxfusion.ai/) is the right choice when your bottleneck is coordinated, high-volume creative production and you can supply a strong brand kit, approved claims, and a testing system. It's the wrong choice if your team hasn't defined the audience, offer, conversion event, or decision rule for moving budget.

## Real Benefits and Real Risks Side by Side

Generative AI reduces the effort required to explore creative directions. This can be problematic when teams confuse exploration with validation.

The business case is strongest for products with frequent creative refresh needs, several meaningful audience contexts, and enough conversion activity to compare variants. It's weaker when a brand has a narrow regulated claim set, minimal traffic, or no reliable conversion instrumentation. In those cases, more assets can create review burden without producing clearer decisions.

Dimension
Real Benefit
Real Risk

Creative iteration
More hooks, formats, and edits can be tested quickly
Near-duplicate variants create noise instead of learning

Cost per asset
Teams can reduce repetitive production work
Lower production cost can encourage uncontrolled volume

Personalization
Messages can adapt to legitimate audience needs
Weak segmentation can produce irrelevant or intrusive ads

Localization
Copy, voiceover, and visuals can support niche markets
Translation errors and cultural mismatches can damage trust

Brand control
Templates and approved references can standardize output
Models may introduce inconsistent details or unsupported claims

Media optimization
AI can help process signals and recommend actions
Platform signals can mask weak creative or low-quality demand

### The controls that make the trade-off work

Start with a **brand kit** that includes approved logos, colors, typography, product references, prohibited claims, tone guidance, and examples of acceptable visual treatment. Add a claim ledger. Every performance promise should have a source, owner, and review status.

Require human review for anything that affects customer understanding. That includes pricing, product capability, testimonials, before-and-after imagery, health or financial implications, synthetic people, voice cloning, and claims about competitors.

Protect intellectual property by using licensed or owned inputs and recording the origin of source material. For AI actors and cloned voices, obtain consent and preserve documentation. A fast workflow without provenance records creates a future audit problem.

Consumer acceptance is conditional. A [2026 U.S. consumer survey summarized by Basis](https://basis.com/news/ai-becomes-standard-in-agency-workflows-with-near-full-industry-adoption) reported that **79% of adults had seen ads that looked AI-made**, **49% viewed business use of AI in advertising negatively**, and **62% considered the practice unacceptable even when disclosed**. The same source described separate research in which **68% of consumers didn't mind AI in ads when it made them more useful or relevant**.

That split should change your creative brief. Don't lead with novelty. Lead with usefulness, clarity, and relevance. Disclosure policies still matter, but disclosure can't rescue an ad that feels generic or deceptive.

## A Practical Implementation Workflow for Startups and Marketers

A working AI advertising system starts with measurement, not a prompt library. Before generating anything, identify the conversion event that matters, the audience quality signal that predicts it, and the baseline campaign you'll compare against.

### Build the operating foundation

Begin with a creative audit. Catalogue existing ads by hook, promise, format, audience, offer, landing page, and outcome. Mark which claims have evidence and which assets are approved for reuse. Then create a compact brand kit that a model can follow without guessing.

Next, build reusable prompt and asset templates. A good prompt includes the audience situation, single message, desired action, product facts, visual constraints, prohibited elements, format, and review requirements. Store the prompt with the resulting asset and its hypothesis, not just the final export.

![A five-step flowchart illustrating a practical implementation workflow for startups using generative AI for advertising strategies.](https://cdnimg.co/b866be35-93f2-4b64-91bc-8c253a419ab8/ef7174d7-9a73-479a-a3e6-879b05e61e94/generative-ai-for-advertising-startup-workflow.jpg)

### Design tests that can teach you something

Create a matrix across **hook, format, audience, and offer**, but avoid changing every variable at once. Each test cell should answer one question. For example, compare a problem-led hook with an outcome-led hook while holding the audience, landing page, and offer constant.

Use image tools for static concept exploration, video tools for demonstrations and UGC-style narratives, and language models for scripts, headlines, and message adaptation. Keep the model responsible for variation, not truth. A human should approve the claims and the final customer interpretation.

A durable launch workflow keeps winning creative discoverable after the paid campaign ends. Publishing a structured launch story, reusable social assets, and an indexable project page through [Aura++ launch content tools](https://auraplusplus.com/projects/makepostai-ai-social-media-content-generator) can give a startup a persistent home for the product narrative instead of leaving every learning inside an ad-platform dashboard. Treat that page as a distribution and documentation layer, not as proof of paid-media lift.

### Close the feedback loop

Attach performance data to the exact creative hypothesis. Record qualified conversion rate, customer quality, cost efficiency, landing-page behavior, and any downstream sales signal available to the team. Feed those findings into the next prompt library revision.

A realistic quarter can follow this rhythm:

- Week one: Audit assets, define the primary business outcome, and establish review rules.
- Week two: Build templates, produce an initial batch, and label every asset by hypothesis.
- Week three: Launch controlled tests with fixed budgets and clear stopping rules.
- Week four: Review quality and incrementality signals, document learnings, and refresh the next batch.
- Later weeks: Expand only the concepts that survive both creative review and business scrutiny.

The common assumption is that AI creates value by making the team faster. In practice, the durable advantage comes from making learning faster without lowering standards.

## Optimization and Measurement Tactics That Separate Winners

Ten variants don't automatically beat one variant. They beat one variant only when the team knows what changed, measures the right outcome, and gives each hypothesis enough exposure to make a decision.

### Test one meaningful idea at a time

Separate message testing from execution testing. If you change the hook, actor, offer, music, landing page, and audience simultaneously, the result can't tell you what caused the difference. Use a test cell built around one primary variable and keep the rest stable enough for interpretation.

Don't rotate creative because a dashboard feels quiet. Define a decision rule in advance, then examine the result against qualified business outcomes rather than clicks alone. The [NYU Stern and Emory research on AI advertising](https://www.stern.nyu.edu/experience-stern/faculty-research/ai-advertising-paradox) reported that fully AI-generated visual ads outperformed human expert-crafted ads by **up to 19% in CTR**, while AI-modified human ads didn't significantly improve performance. It also reported that explicitly disclosing AI use reduced CTR by **31.5%** in the studied setting.

That finding isn't a universal prescription to hide AI use. It is a warning that execution and perception affect results, and that disclosure strategy needs legal, platform, and audience review.

### Improve the signal before improving the model

Feed qualified-account behavior back into audience seeds instead of optimizing only for cheap clicks. If the platform receives a noisy conversion signal, it will find more people who resemble the noisy signal. Your creative testing framework can't compensate for poor event definitions.

Use holdout groups when the campaign and channel allow it. For offline or geographically concentrated influence, consider geo-based lift testing. Attribution reports can support diagnosis, but they shouldn't be treated as proof of incrementality without a comparison design.

Tactic
Vanity Metric It Looks Like It Moves
Business Metric It Actually Moves
When To Trust It

More creative variants
Asset count and output speed
Learning rate and qualified conversion efficiency
When each variant represents a distinct hypothesis

Hook testing
CTR
Qualified conversion rate and downstream value
When the landing experience stays consistent

Audience refinement
Reach and CPC
Lead quality and customer acquisition efficiency
When conversion events reflect real business value

Landing-page alignment
Bounce rate
Signup completion or purchase intent
When traffic source and offer remain comparable

Holdout testing
Reported platform conversions
Incremental conversions
When exposure and comparison groups are clean

Durable content publishing
Referral clicks
Search visibility and reusable discovery
When pages are structured, indexed, and maintained

For teams managing Google campaigns, an [AI-assisted Google Ads intelligence workflow](https://auraplusplus.com/projects/adsoar-google-ads-intelligence) can help organize campaign signals and decisions. Keep the role clear. Intelligence tooling should improve analysis and action, not become another dashboard that rewards activity.

The UK advertiser evidence exposes the measurement gap directly. [ISBA's survey on effective advertising and generative AI](https://www.isba.org.uk/article/major-survey-reveals-impact-gen-ai-effective-advertising) reported that **99% of UK advertisers were engaging with generative AI**, yet only **14% reported significant impact on business results**. Among advertisers prioritizing effectiveness, the share reporting major impact rose to **55%**, and those advertisers were twice as likely to say generative AI had meaningfully improved results. The same source reported **69% ROI for advertising and media firms**, compared with a **49% cross-industry average**.

The conclusion is direct. Measurement discipline, not asset volume, separates useful adoption from expensive experimentation.

## Your First 30 Days and What to Watch Next

A small team can establish a credible AI advertising workflow in four weeks if it limits the scope. Choose one campaign, one primary outcome, and a manageable set of creative hypotheses. Don't attempt to automate the entire marketing function before you know which loop deserves automation.

![A four-week infographic roadmap illustrating a structured action plan for implementing generative AI in advertising campaigns.](https://cdnimg.co/b866be35-93f2-4b64-91bc-8c253a419ab8/b4e48c1b-d55f-422e-a44f-19a2e7ac1470/generative-ai-for-advertising-action-plan.jpg)

### Week one, audit and goal setting

Inventory current assets and identify the campaign's baseline. Choose the business metric that decides success, then define supporting indicators such as qualified conversion rate, landing-page completion, sales acceptance, or retention. Write down what would cause you to pause, revise, or scale a test.

### Week two, prompt library and creative sprint

Draft core prompts for your main audience, offer, formats, and brand voice. Generate a focused batch of images, scripts, and video variations. Review claims, visual continuity, consent, disclosure requirements, and accessibility before anything reaches the ad account.

### Week three, structured testing

Launch tests with one principal variable per cell. Keep budgets, landing pages, and conversion definitions stable enough to compare outcomes. Watch frequency, audience quality, comments, lead validation, and post-click behavior, not only the platform's preferred engagement metric.

### Week four, review and scale

Separate winners from interesting failures. A winning ad should have a clear reason for success, a repeatable message, and evidence that the resulting action matters. Document the prompt, source assets, audience, placement, and result so the next production cycle starts with learning rather than a blank page.

Use durable distribution for the product narrative and launch assets. Paid ads create exposure, while structured, searchable pages can preserve the positioning, creative context, and product information for later discovery. That combination is more resilient than relying on a one-time spike inside a social feed.

Watch the next phase closely, but don't chase every announcement. AI-generated video, synthetic voice, stronger disclosure and labeling requirements, model commoditization, and agentic optimization workflows will change production economics. They won't remove the need for a clean offer, credible claims, qualified signals, and controlled measurement.

Make these decisions this week:

- Choose the outcome: Decide which business result outranks impressions and clicks.
- Name the hypothesis: Write the customer problem and message each asset will test.
- Set the guardrails: Approve claims, references, actors, voices, and review ownership.
- Define the comparison: Choose the baseline, holdout, or lift method before launch.
- Plan the durable layer: Decide where the product story and reusable assets will remain discoverable.

Generative AI for advertising is already practical. The teams that benefit won't be the ones generating the most creative. They'll be the ones that connect useful creative to clean signals, prove incremental value, and preserve the learning in a workflow they can run again.

This week, audit one active campaign, define its business outcome, and create a small test matrix with documented hypotheses. Then publish the approved launch narrative and reusable creative assets through a durable distribution channel such as Aura++, so your campaign work can support both immediate testing and ongoing product discovery.

## Links

- Article: https://auraplusplus.com/blog/generative-ai-for-advertising
- AI-friendly Markdown: https://auraplusplus.com/blog/generative-ai-for-advertising.md
- Blog index: https://auraplusplus.com/blog

## Featured image

- https://cdnimg.co/b866be35-93f2-4b64-91bc-8c253a419ab8/7cd45da9-e21a-476a-b002-6c75f7e2042d/generative-ai-for-advertising-presentation-title.jpg

## Explore

- [More on the Aura++ blog](https://auraplusplus.com/blog)
- Category: generative ai for advertising


_HTML version: https://auraplusplus.com/blog/generative-ai-for-advertising_