AI in Marketing Takeaways for Leaders: August 2026

Written By Sam Gadsby

What happens when AI has the context, access and authority to act?

For much of the past few years, business use of AI has been relatively contained. Ask a question. Generate some content. Analyse a document. Get an answer. Decide what to do next. That boundary is now rapidly disappearing.

AI now increasingly sits inside email, documents, spreadsheets, advertising platforms and business systems. It can understand more context, use tools, create workflows and take actions on our behalf. At the same time, models are becoming cheaper to run, easier to deploy and available to far more people.

That combination creates significant opportunity for marketing and commercial teams. It also changes the leadership challenge.

What happens when AI has enough context to understand the situation, enough access to do something about it and enough authority to pursue an outcome?

Several developments over the past month give us a useful view of where this is heading.

1. The ‘rogue AI’ headlines contain an important lesson about goals and permissions

One of the more striking AI stories this month involved OpenAI and Hugging Face. During an internal cyber-capability evaluation, OpenAI models were tasked with solving security challenges inside a deliberately constrained testing environment.

Rather than following the expected route, the models found and exploited a previously unknown vulnerability to gain internet access. They then compromised Hugging Face infrastructure while attempting to retrieve answers to the evaluation.

OpenAI described it as an “unprecedented cyber incident”. Its investigation found the models had become highly focused on achieving the narrow objective they had been given, going to considerable lengths to find another route to the answer.

Reuters subsequently reported three similar containment incidents disclosed by Anthropic. The phrase “going rogue” has inevitably appeared around these stories, although researchers have rightly cautioned against suggesting the systems independently developed malicious intent.

The impact an AI system can have increasingly depends on the combination of capability, objective and permission. A marketing assistant that can recommend a campaign change creates one level of risk. One that can alter budgets creates another. The same applies across CRM, email, customer service, pricing and commercial systems.

A useful way to think about authority is: Read → Recommend → Create → Change → Transact

Every movement along that scale should come with an intentional decision about oversight.

Leadership takeaway:

As AI agents become more capable, governance needs to move beyond acceptable-use policies.

For important workflows, be clear about:

  • What information the AI can access

  • Which actions it can take

  • Where human approval remains required

  • How activity is logged and monitored

  • What happens when behaviour falls outside expectations

Good AI governance is increasingly part of good operational design.

2. AI is becoming part of the software estate, and that may be helping Gemini close the gap

The competitive AI market is becoming much more interesting. Similarweb's latest data shows ChatGPT's share of global generative AI web traffic falling from around 76% a year ago to approximately 53%, while Gemini has increased to around 27–28%. Google's July results also reported 950 million monthly active Gemini users, with daily active usage tripling over the previous year.

There will be several reasons for that growth, but Google's distribution advantage is hard to ignore. Gemini increasingly connects across Search, Workspace, Gmail, Docs, Sheets, Android, YouTube and Google's wider product ecosystem. This month gives us another good example.

Google launched Sheets canvas on 13 August, allowing users to describe what they want and turn spreadsheet data into interactive mini-apps without coding. The resulting interface stays connected to the underlying spreadsheet and can be edited through further prompts.

Google's new Gemini 3.7 Flash also strengthens its agentic capabilities across Workspace, while Gemini Spark can consolidate files, draft emails and update status documents as part of multi-step workflows.

And Google won't necessarily own this layer simply because a business uses Workspace. xAI launched Grok directly inside Google Workspace and Microsoft Outlook during the same period, giving users another AI assistant within software they already use.

The strategic question therefore becomes broader than which chatbot scores highest on a benchmark.

Leadership takeaway:

When reviewing AI platforms, consider how well they connect with the environment people already work within. Model quality matters, but so do:

  • Access to relevant company information

  • Integration with everyday workflows

  • Security and permission controls

  • Ease of adoption

  • Overall licence and technology cost

For many organisations, the strongest AI platform may ultimately be the one people can use naturally within their existing work.

3. AI is getting cheaper, but AI budgets may still get bigger

The economics of AI are moving quickly. OpenAI reduced the API price of GPT-5.6 Luna by 80% at the end of July and Terra by 20%, arguing that increasingly capable models can now handle high-volume business workloads much more economically.

Google followed with Gemini 3.7 Flash on 13 August. Its introductory pricing is half the original Gemini 3.6 Flash price, alongside significant improvements in coding, knowledge work and agentic workflows.

That sounds like straightforward good news for AI budgets. I wouldn't assume it means AI becomes a smaller cost line.

Cheaper intelligence makes many more use cases commercially viable. Businesses that once used AI occasionally may run thousands of automated tasks. Teams may add multiple specialist platforms alongside core ChatGPT, Gemini or Copilot licences.

There is also a useful reminder within Google's announcement: its introductory Gemini 3.7 Flash pricing expires on 31 December 2026, with higher pricing already stated for January. The broader cost base can include:

  • User licences and specialist AI tools

  • API and automated-agent usage

  • Integration and data infrastructure

  • Security and governance

  • Implementation and ongoing optimisation

That matters particularly for organisations building an AI business case around headcount or supplier savings. If team structures, agency requirements or production costs reduce, some of that saving may reappear elsewhere as technology consumption grows.

Leadership takeaway:

Manage AI investment against value created, rather than simply licence cost or cost per token.

Not everybody needs the most capable model or highest subscription tier. Review the use case, frequency, required quality, cost of error and expected commercial benefit. Then use the right level of capability for the job. AI should still have to earn its budget.

4. AI advertising creates a new commercial relationship with customer context

ChatGPT advertising is evolving quickly from an interesting experiment into something that increasingly resembles a performance-media platform. OpenAI's Ads Manager Beta now supports campaign management and performance reporting, while its advertising guidance makes clear that relevance is based heavily on conversational intent rather than conventional keyword matching.

There is a particularly interesting difference from established advertising environments. When personalisation is enabled, OpenAI says ad relevance can use signals from the current conversation and, depending on user settings, previous chats and memory. Advertisers themselves do not receive those conversations or memories.

From a marketer's perspective, that context could be incredibly valuable.

People often tell AI assistants much more about the problem they are trying to solve than they would ever put into a traditional search query. The industry is already pushing the idea further.

Time has begun testing advertisements designed specifically for AI crawlers rather than human readers, with brands including Ally Bank and the Project Management Institute among the first advertisers. The aim is to provide information that could influence how AI systems understand and subsequently describe those brands. Perplexity has since pushed back on the approach, describing bot-only content as a form of cloaking.

It is early, experimental and potentially messy. But it raises a question marketers are going to hear much more often:

When an AI system helps someone research and choose a product, who are we actually marketing to: the customer, the assistant, or both?

For more on the channel and search implications, see this month's Search Marketing Takeaways for Leaders: August 2026.

Leadership takeaway:

AI advertising deserves experimentation, but trust should remain central.

The opportunity comes from understanding customer intent and being genuinely relevant within that context. Attempts to manipulate the intermediary are far more likely to create a short-term arms race than sustainable competitive advantage.

5. AI is redrawing jobs faster than most organisations are redrawing responsibilities

Some of the most interesting AI workforce research this month is about what people are actually doing differently.

OpenAI analysed more than 800,000 work-related ChatGPT messages and found that 43.5% of occupation-specific usage involved tasks associated with another occupation.

Marketing stood out in both directions. Marketers were using AI to undertake work traditionally associated with other functions, while people elsewhere in organisations were increasingly undertaking marketing-related tasks themselves.

That feels much closer to the reality many businesses are experiencing than the simple debate about whether AI replaces jobs.

  • A marketer can analyse a dataset without waiting for an analyst.

  • A salesperson can produce first-draft campaign content.

  • A commercial leader can interrogate financial information.

  • Someone without development skills can now create a simple internal application from a spreadsheet.

The boundaries around roles become less rigid.

For leaner businesses this can be particularly powerful. Fewer handoffs can mean quicker decisions and less reliance on specialist resource for relatively straightforward work.

There is a leadership challenge alongside that flexibility. Expertise still matters. Someone being able to perform a task does not automatically mean they can recognise whether the output is good. As generation becomes easier, judgement becomes more valuable.

That applies to creative work too. The ability to produce acceptable copy, imagery, presentations or video is rapidly becoming commonplace. Strong ideas, customer insight, brand understanding, direction and editing become more important when production itself is abundant.

Leadership takeaway:

Review roles around capabilities and accountability, rather than simply asking which tasks AI can automate.

Consider:

  • Which capabilities should become more widely distributed

  • Where specialist expertise remains essential

  • Which repetitive handoffs can disappear

  • Who remains accountable for quality

  • How people develop judgement when AI does more of the production

AI workforce planning and AI technology planning increasingly belong in the same conversation.

6. AI transparency is becoming an operational marketing issue

A significant milestone arrived on 2 August when transparency obligations under Article 50 of the EU AI Act began to apply.

The requirements cover areas including informing people when they are interacting with certain AI systems and ensuring some AI-generated or manipulated content can be identified in machine-readable form. Specific disclosure rules also apply to deepfakes and certain public-interest content.

There are important nuances and exceptions, including where AI performs an assistive editing function without substantially altering the original material. This is an area where organisations should take appropriate legal advice rather than rely on a generic interpretation.

For marketers, the practical challenge is already visible. Canva formally classifies tools including Background Remover as AI Products. Its current terms also refer to provenance and metadata around AI-generated content.

Meta has separately acknowledged that relatively minor AI-assisted edits have previously triggered its AI labelling systems because editing software can include industry-standard provenance signals. It consequently changed how it displays its “AI info” labels for content that has been edited rather than fully generated.

That doesn’t mean using Canva's Background Remover will automatically result in an AI label on Instagram or Facebook.

It does demonstrate how quickly definitions can become complicated. A marketer may think an image has simply been edited. A creative platform may classify the feature as AI. Another platform may detect AI-related metadata. Regulation may treat assistive editing differently again.

Leadership takeaway:

Build AI provenance into normal marketing governance. Teams should know:

  • Which tools are using AI

  • How important assets were created or edited

  • When disclosure requirements may apply

  • Whether agencies and partners follow the same standards

  • Who owns the final compliance decision

Transparency is becoming part of content operations, not an issue to solve after publication.

7. More AI agents could simply create faster organisational complexity

Agentic AI is one of the strongest themes across current product development. Google's Gemini 3.7 Flash is explicitly designed for agents and multi-step workflows, with Google's own examples showing the model orchestrating sub-agents to complete more complex work.

There is obvious potential here. Different agents could research, analyse, create, monitor and execute parts of a workflow simultaneously. But businesses have spent decades trying to solve another problem: too many disconnected teams and systems optimising their own objectives. It would be fairly easy to recreate exactly the same issue with AI.

Imagine a marketing agent optimising lead volume, a sales agent prioritising conversion, a finance agent reducing acquisition cost and a customer agent protecting satisfaction.

Each could make perfectly logical decisions in isolation while collectively pulling the business in different directions. Speed doesn't remove the need for alignment.

Leadership takeaway:

Before multiplying agents, make sure the underlying objective is clear.

Start with the commercial outcome, define ownership and understand how different automated decisions interact.

Otherwise we risk automating organisational complexity rather than removing it.

8. AI will change customer expectations as well as employee productivity

Most AI strategies still begin internally.

  • How can we produce content faster?

  • How can we analyse information?

  • How can we automate administration?

  • How can teams become more productive?

All worthwhile questions. The customer side may ultimately be just as important.

People are becoming accustomed to systems that can understand natural language, retain context, compare options and move from answering a question towards taking action.

OpenAI is explicitly positioning ChatGPT advertising around users who can explore, compare and decide within one conversational experience. Google's Gemini agents are increasingly designed to move from information towards action across connected services.

Those experiences gradually change the benchmark customers bring to every other digital interaction.

  • A website search that requires exactly the right keywords can feel less useful.

  • A form that repeatedly asks for information the business already has creates more friction.

  • A customer-service journey that forces somebody to explain the same problem to several departments feels increasingly dated.

Leadership takeaway:

Include customer experience within your AI strategy.

Look at the points where customers currently have to:

  • Search through complexity

  • Repeat information

  • Compare multiple options themselves

  • Wait for routine answers

  • Move unnecessarily between channels

Some of the highest-value AI opportunities may come from making those experiences easier rather than simply making marketing production quicker.

Leadership Perspective: AI is becoming part of the operating model

The developments this month are connected. AI is getting cheaper. It is gaining access to more information. It is moving into everyday software. It can undertake work that previously required specialist skills.

Agents can increasingly act rather than simply advise. Advertising models are emerging around the context people share with assistants. Regulation and platforms are starting to formalise how AI involvement should be disclosed.

For marketing and commercial leaders, this takes AI beyond experimentation. The quality of the underlying operating model starts to matter. Five areas deserve particular attention:

  • Purpose: What valuable business or customer outcome are we trying to improve?

  • Access: What information and systems should AI be allowed to use?

  • Authority: What can it recommend, create, change or complete?

  • Economics: What value is being created relative to the full cost?

  • Accountability: Who remains responsible for quality, risk and the final outcome?

Those questions should stay fairly stable even as the models and platforms continue to change.

Leader's Checklist

A few useful questions to ask this month:

  • Do we know which AI systems have access to important company or customer information?

  • Are AI licences and usage being reviewed against real use cases and value created?

  • Have marketing, people and technology teams discussed how AI is changing role boundaries?

  • Do our content and creative processes have a sensible approach to AI provenance and disclosure?

  • Are we considering how AI could improve the customer experience as well as internal efficiency?

Final Thought

The first phase of business AI was largely about output: Can it write this? Summarise that? Analyse these numbers? Create this image?

The next phase looks much more consequential. AI increasingly has context about the situation, access to the systems involved and the ability to take action. That can make organisations faster, more capable and potentially much more efficient. It also means AI decisions become business decisions.

Marketing and commercial leaders therefore need to stay close to how these systems are deployed, what they cost, what they can access and where responsibility ultimately sits. The organisations that get the most from AI are unlikely to be those adopting every new capability first.

They will be the ones that connect capable technology to clear commercial priorities, give it appropriate authority and retain enough human judgement around the decisions that matter.

Useful Links & Further Reading

Sam Gadsby
Next

Search Marketing Takeaways for Leaders: August 2026