Welcome. This course will prepare you to analyse how technology companies define and communicate “ethical AI,” safety, responsibility, and risk—and to assess when those communications may function as legitimation or ethics-washing. This module focuses on strategic framing: the communicative choices through which a company makes one interpretation of AI seem natural, urgent, credible, or responsible.
A corporate frame is more than a positive statement about values. It is an organising interpretation: it makes certain problems, causes, moral priorities, and solutions prominent, while placing other interpretations in the background. In this lesson, you will learn to identify the smaller linguistic and narrative elements that reinforce such a frame: keywords, metaphors, contrasts, and roles assigned to companies, users, experts, regulators, and AI systems.
From a corporate sentence to a corporate frame
Framing is not necessarily deception. Any organisation communicating about a complex topic must select language, examples, sources, and priorities. The analytical task is therefore not to ask whether a text “has a frame.” It always does. Instead, ask:
- What interpretation of AI is being made salient?
- Which language choices make that interpretation coherent?
- Which actors are presented as responsible, vulnerable, authoritative, or peripheral?
- What alternative interpretations become less visible?
For example, consider this illustrative corporate statement:
“Advanced AI can transform public services, but its benefits depend on rigorous safeguards. We are committed to building a safety-first platform, working with independent experts so that organisations can deploy AI with confidence.”
This is not simply a collection of claims. Taken together, it constructs a frame in which:
- AI is a beneficial force for public transformation;
- risks are real but manageable through safeguards;
- the company is a committed safety engineer;
- independent experts provide validation;
- organisations are responsible adopters who can proceed confidently.
The core frame might be summarised as: AI progress is legitimate when managed through company-led technical safety and expert oversight.
The individual words do not prove that frame by themselves. Their value lies in how they work together.
The Framing Theory in Media and Communication
Watch The Framing Theory in Media and Communication by Ask Prof. Rachel for a concise grounding in framing as selection, emphasis, and omission. Although the examples concern media, the logic transfers directly to corporate webpages, policy papers, safety updates, and product announcements.
Watch packaging meaning to see how vocabulary, visuals, expressions, and presentation choices position an issue within an interpretive context. Then watch selection and exclusion, especially the discussion of terminology: a label does not merely describe an issue but can guide how audiences interpret it.
A useful distinction for your thesis is:
| Level | What you identify | Example |
|---|---|---|
| Textual marker | A word, phrase, image, contrast, or attribution | “safety-first,” “but,” “independent experts” |
| Immediate function | What that marker does locally | Presents safety as a priority; pivots from risk to solution; adds external credibility |
| Frame | The broader interpretation supported by multiple markers | AI can be advanced responsibly under company-led safeguards |
Do not jump directly from one word to an accusation such as “ethics-washing.” A word such as trustworthy is an observation. Its framing role depends on its context, repetition, accompanying evidence, and relation to what is not said.
Four families of framing devices
Gamson and Modigliani describe a frame as a central organising idea supported by “condensing symbols”: recurring devices that make a broad interpretation memorable and plausible. For corporate AI communication, four particularly useful families are keywords, metaphors, contrasts, and narrative roles.
Understanding the Models of Framing Analyses Approaches in Media
Read this short overview of Gamson and Modigliani’s “media package” approach. It gives you a practical vocabulary for distinguishing a core frame from the textual and visual devices that support it.
In Section 2.1, “Internal Structure of Media Package” (pp. 385–386), begin immediately after the discussion of the package’s “core frame.” Read the account of framing and reasoning devices. Focus on the distinction between the central organising idea and its supporting metaphors, exemplars, catchphrases, depictions, visual images, roots, consequences, and appeals to principles. Translate “news text” into your future corpus of corporate AI materials.
1. Keywords: recurring evaluative vocabulary
Keywords are terms whose repeated use clusters an issue around particular values, priorities, and assumptions. They may be nouns, adjectives, verbs, or short formulaic phrases. In corporate AI communication, common clusters include:
| Lexical cluster | Possible examples | Likely framing implication |
|---|---|---|
| Innovation and progress | frontier, breakthrough, unlock, transform, empower, scale | AI is a source of advancement or social benefit |
| Safety and control | safeguards, guardrails, robust, secure, aligned, red lines | Risk is manageable through designed controls |
| Ethics and social value | fairness, inclusion, human-centred, trustworthy, responsible | The company connects its work to moral values |
| Governance and process | oversight, assessment, review, standards, accountability | Responsibility is presented as an organisational process |
| Uncertainty and limitation | may, could, evolving, complex, emerging | The company preserves flexibility or acknowledges uncertainty |
| Collaboration and authority | experts, researchers, partners, stakeholders, independent | Legitimacy is supported by association with other actors |
The aim is not to make a universal list and mechanically count terms. The same word can work differently in different contexts.
Compare these two statements:
- “Our systems are safe because they undergo independent evaluation before release.”
- “We are committed to safe AI.”
Both invoke safety. The first makes a potentially checkable claim about a process and assigns a role to an evaluator. The second makes safety part of the company’s public identity but does not specify what safety means, who judges it, or what happens when a system fails.
When annotating a keyword, record its co-text: at least the clause or sentence around it. A standalone frequency count of responsible, ethical, or safe cannot establish what the company means by those terms.
2. Metaphors and catchphrases: importing a familiar story
A metaphor links a difficult or abstract target domain—such as AI governance—to a more familiar source domain. It is not decorative. It can imply what the problem is, who has agency, and what kind of response appears sensible.
| Corporate expression | Source domain | Implied interpretation of AI responsibility |
|---|---|---|
| “AI guardrails” | Roads or railways | Innovation should continue, but within engineered limits |
| “The AI frontier” | Exploration and territorial expansion | AI development is pioneering, uncertain, and difficult to halt |
| “A responsible AI ecosystem” | Natural interdependence | Responsibility is shared across many actors, potentially diffusing power |
| “The global AI race” | Competition or conflict | Speed and national or commercial competitiveness become pressing priorities |
| “A safety stack” | Engineering architecture | Ethical responsibility is organised as layered technical controls |
| “Building trust” | Social relationships | The company must be seen as reliable, not merely legally compliant |
Take “guardrails.” It does more than say “we have rules.” It suggests that movement is expected and desirable; the task is to keep that movement on a safe path rather than question its destination. This can support a frame in which the main ethical challenge is technical risk management, rather than, for example, whether a particular application should exist or who holds power over it.
A useful diagnostic is to ask what would sound strange if the metaphor were taken literally. A “frontier” has explorers, unknown territory, and hazards. A “race” has competitors, pressure to accelerate, and winners and losers. These implied elements may organise the corporate story even when they are not stated explicitly.
Not every vivid word is a metaphor. Terms such as model, alignment, or robustness may be technical terms in a particular context. Code them as metaphors only when the wording imports an external conceptual domain and appears to shape the interpretation of responsibility.
3. Contrasts: defining ethics by drawing a boundary
Contrasts tell the reader what the company wants to reject, qualify, or supersede. They are particularly valuable because they reveal the boundaries of a corporate definition of “ethical AI.”
Look for linguistic forms such as:
- but, however, while, although, despite
- not just … but …
- rather than
- beyond compliance
- not a replacement for
- cannot guarantee
- may not
Consider:
“Trustworthy AI requires considering not just what data is legally available to use, but what data is socially responsible to use.”
The contrast establishes legal compliance as insufficient and elevates social responsibility as the preferred standard. This is a meaningful expansion of responsibility—at least rhetorically. It tells us that the company wishes to be associated with a standard above bare legality.
Now compare:
“Our safeguards reduce risks; however, no system can guarantee complete security.”
Here, the contrast performs a different function. It foregrounds responsible effort, then limits the scope of the promise. That can be appropriately transparent, especially where absolute safety truly cannot be guaranteed. Yet it also manages expectations and potential liability. Your analysis should preserve both possibilities rather than treating cautious language automatically as evasive.
The key question is: Which side of the contrast receives the moral and rhetorical weight? In “not just legal, but socially responsible,” social responsibility is privileged. In “risks exist, but innovation must continue,” continuation may be privileged over precaution.
4. Narrative roles: who acts, who benefits, who bears risk?
Frames are stories in a minimal sense: they allocate roles. Corporate communication often casts the organisation in a favourable role—innovator, steward, protector, partner, or responsible governor—while distributing different roles to others.
Common roles in AI ethics communication include:
| Role | Typical textual signals | Analytical significance |
|---|---|---|
| Company as protagonist | “We lead,” “we build,” “we commit,” “our approach” | The company is presented as the primary ethical agent |
| Company as protector or steward | “protect users,” “safeguard society,” “earn trust” | The company assumes a guardianship role |
| AI as tool | “AI assists,” “AI enables,” “AI supports” | Human or organisational control is emphasised |
| AI as force or actor | “AI transforms,” “AI disrupts,” “AI threatens” | Technology may appear autonomous, reducing attention to corporate choices |
| Users as beneficiaries | “empower people,” “help customers,” “improve lives” | Benefits are made visible, often in broad terms |
| Affected people as vulnerable subjects | “protect children,” “prevent discrimination,” “support communities” | Harm and duty of care become more salient |
| Experts as validators | “independent researchers,” “external advisers,” “scientific consensus” | External authority supports credibility |
| Regulators as partners or constraints | “working with policymakers,” “meeting requirements” | Governance may be framed as collaboration or compliance burden |
| Critics as absent or distant voices | “some claim,” “concerns have been raised,” scare quotes | Dissent may be acknowledged without being treated as authoritative |
Pay attention to grammatical agency. “We tested the system for disparate impact” identifies a company action. “Bias may arise in data” makes bias appear almost natural or impersonal. The second sentence does not necessarily deny responsibility, but it directs attention away from choices about data collection, model design, deployment incentives, and organisational accountability.
Voice positioning: how a company handles agreement and disagreement
Narrative roles show who occupies the corporate story. Engagement markers show how the corporate voice positions itself in relation to other possible voices and viewpoints. This distinction is especially useful when you analyse claims about safety, transparency, or stakeholder consultation.
Engaging stakeholders, shaping AI ethics: Targeted engagement in corporate AI ethics statements
Read the engagement-system framework in this article on corporate AI ethics statements. It provides a usable taxonomy for identifying whether companies assert, hedge, deny, counter, or attribute ethical claims.
In Section 3.1, “The engagement system,” read the paragraph beginning with the opening explanation, then continue through the definitions of Proclaim, Disclaim, and Expansion. In Section 4.2.2, “Identification of engagement strategies,” read the coding setup and study Table 1 in full. Focus on the examples alongside each category; they show why the same topic can be communicated with markedly different degrees of certainty, obligation, and openness to alternatives.
Use the following condensed guide when coding:
| Engagement move | Typical markers | What it does to the frame |
|---|---|---|
| Pronounce | “we are committed,” “we will,” “we affirm” | Presents the corporate position as firm and authoritative |
| Concur | “clearly,” “of course,” “everyone agrees” | Treats a view as common sense or shared knowledge |
| Endorse | “research shows,” “the study found” | Borrows authority from an external source |
| Deny | “do not,” “will not,” “no” | Rejects an undesirable interpretation or practice |
| Counter | “but,” “however,” “while,” “despite” | Acknowledges one view before advancing a preferred one |
| Entertain | “may,” “could,” “we believe,” “it seems” | Leaves room for uncertainty, alternatives, or discretion |
| Acknowledge | “experts stated,” “a report noted” | Attributes a position without clearly endorsing it |
| Distance | “they claim,” scare quotes | Marks an attributed view as questionable or external |
The article’s empirical table lists seven strategies; its theoretical discussion also explains Distance. For your codebook, it is reasonable to retain distance as a possible category if it appears in your corpus, but define it clearly and apply it consistently.
A particularly important pattern is the combination of strong commitment with flexibility:
“We will continue to develop inclusive measures that could improve outcomes.”
“We will continue” is a pronounce marker: it portrays commitment. “Could” is an entertain marker: it lowers certainty about outcomes. Together, the sentence can sound ethically determined while retaining room for non-delivery. That does not establish deception. It does identify a tension worth tracing across documents and against available evidence later in the study.
A worked analysis: assembling the evidence
Return to the illustrative statement:
“Advanced AI can transform public services, but its benefits depend on rigorous safeguards. We are committed to building a safety-first platform, working with independent experts so that organisations can deploy AI with confidence.”
A disciplined annotation might look like this:
| Excerpt | Device | Local function | Contribution to the broader frame |
|---|---|---|---|
| “Advanced AI” | Keyword cluster | Associates AI with technological progress | AI is positioned as valuable and forward-looking |
| “transform public services” | Progress vocabulary | Makes benefits socially significant | Adoption appears publicly beneficial, not merely commercial |
| “but” | Counter | Shifts from promise to qualification | Risk is recognised but presented as compatible with progress |
| “rigorous safeguards” | Safety keyword | Presents risk as governable through controls | Ethical responsibility becomes a matter of safeguards |
| “safety-first platform” | Catchphrase and metaphor | Links safety to engineered infrastructure | The company appears technically capable of managing ethics |
| “We are committed” | Pronounce | Signals resolve | Company is cast as responsible moral agent |
| “independent experts” | Narrative role and endorsement cue | Supplies external validation | Corporate action appears credible and not wholly self-judged |
| “deploy … with confidence” | Outcome language | Normalises continued adoption | Responsible governance is portrayed as enabling, not restricting, deployment |
The image captures an important discipline for your later thesis work: do not treat the frame as a vague impression. Move from observable textual evidence to a documented interpretive claim. In this lesson, the observable evidence includes the vocabulary of transformation and safeguards, the contrast introduced by “but,” the metaphor of a safety-first platform, and the allocation of roles to company and experts.
A practical annotation record can contain five fields:
- Exact excerpt — preserve the wording and URL or document reference.
- Device label — keyword, metaphor, contrast, narrative role, or engagement marker.
- Immediate effect — for example, “casts risk as technically manageable.”
- Frame link — explain which larger interpretation it reinforces.
- Alternative reading or uncertainty — note ambiguity rather than forcing a single interpretation.
For example, “independent experts” may genuinely indicate meaningful external oversight, or it may function primarily as a credibility signal. At this stage, code the role and framing function. Later, you will assess whether the organisation identifies those experts, explains their authority, and provides evidence of their influence.
What careful interpretation looks like
Three safeguards will make your analysis more credible.
First, analyse patterns rather than isolated words. A company saying “ethical,” “responsible,” or “human-centred” once is weak evidence of a stable frame. Repetition across genres, coherence with metaphors and roles, and prominence in headlines or conclusions are stronger evidence.
Second, distinguish textual interpretation from claims about intent. You can write: “The statement frames safety as an engineering challenge under corporate control.” You should not write: “The company deliberately manipulates readers” unless you have evidence that can support such a claim. Framing analysis is strongest when it stays close to what the communication demonstrably does.
Third, attend to the boundary being constructed. If a text calls AI ethical because it is safe, ask what “safe” covers. Security? Model misuse? Discrimination? Labour conditions? Military use? Environmental costs? A frame is often clearest not only in what it includes but in how it narrows the relevant ethical field.
Key takeaways
A corporate frame is reinforced through a coordinated set of small communicative choices:
- Keywords associate AI with selected values, risks, and priorities.
- Metaphors and catchphrases import a familiar logic, such as control through guardrails or urgency through a race.
- Contrasts draw boundaries around what counts as responsible, sufficient, or realistic.
- Narrative roles assign agency, authority, vulnerability, and benefit to particular actors.
- Engagement markers reveal whether the company asserts, hedges, rejects, attributes, or distances itself from alternative viewpoints.
Your task is to connect these observable devices to a defensible statement about the interpretation of AI that a text makes salient—without prematurely treating framing as evidence of ethics-washing or hidden intent.
In the next lesson, you will classify corporate communication genres—such as safety reports, product pages, policy submissions, blog posts, and public principles—and consider how genre shapes what companies can credibly claim about ethical AI.
Can't find a good explanation? Sign up and we'll make it for you
Sign up