Afternoon BriefAI Search & Discovery

Sentiment Delta in Brand AI Search: What Seer's 84-Session Study Shows About AI and Purchase Consideration

Seer Interactive's study of 28 participants and 84 AI research sessions found one AI session could cut a brand's purchase consideration sharply. How to record your own sentiment delta and trace its source.

Christian Lehman
Christian LehmanJul 7, 2026

Seer Interactive interviewed 28 people, ran 84 AI-assisted research sessions, and asked each participant which brands they would consider before and after using AI to pick a product. Seer reports that the number of participants who considered GE fell from 18 to 8 and Whirlpool from 9 to 2, while Rheem, a lesser-known water heater brand, gained. Seer's own summary is that one AI research session cut GE's purchase consideration in half. It is a small, consumer-product study, not a market measurement, and it is a vivid illustration of a gap most brand tracking does not look at.

What the Seer Interactive Study Measured

The Seer study design was straightforward: ask participants which brands they would consider before AI research, have them use AI to find a product they would actually purchase, then ask which brand they now intend to buy and why. 28 participants, three product tasks each, 84 AI research sessions.

The tasks covered products such as appliances, faucets and water heaters. Seer reports GE dropping from 18 to 8 and Whirlpool from 9 to 2 among participants who considered them, Samsung holding up better than GE in the same appliance categories, and Delta (faucets) and Rheem (water heaters) gaining.

Two cautions. The sample is small and the tasks were consumer purchases, so this is evidence that AI research can reshuffle a consideration set, not an estimate of how much, for which brands, or in B2B. And Seer's design measures stated purchase intent, not purchases.

AI answers are also unstable between phrasings. Research on paraphrase brittleness found that two natural paraphrases of the same buying intent produced recommendation sets overlapping only 14–29% (Jaccard), against 50–61% for reruns of the same prompt. The brand a buyer sees depends partly on how they ask.

That is the measurement gap. A brand tracker can show stable awareness and an SEO dashboard steady rankings while buyers use a channel that changes who gets considered.

Sentiment Delta Is Not a Score. It Is a Recorded Gap

In Machine Relations, sentiment delta is the recorded difference between a brand's declared positioning and the descriptions AI systems return for a fixed query set. It is an observation metric. It does not by itself say why the description is what it is, or whether it changed buyer behavior. AuthorityTech uses the term in its Machine Relations stack.

That distinction matters because most AI sentiment monitoring tools, including Profound, Loamly and SEOcrawl as Attrifast describes them, tell you what the model says now. They do not tell you which source drove the wording or what it costs.

Attrifast's revenue analysis overlaid sentiment on a subset of its roughly 200 sites and found lower conversion after negative AI answers, concentrated on evaluative and comparison queries, with the author stating that causation is not proven. A stale review is a different problem from a missing source page on a comparison query, and the fix differs. The score alone does not distinguish them.

The Four Layers of AI Sentiment Formation

Searchbloom's Cody Jensen describes sentiment as a four-layer system rather than one number:

  • Layer 1: Brand awareness, what a human asker already feels about the brand before typing the prompt.
  • Layer 2: LLM parametric memory, what the model knows from training, which Searchbloom says changes only at training cutoffs. Stale negatives here, like the founder in Attrifast's case study whose resolved downtime incident was still presented as present-tense fact, can persist until the next retraining.
  • Layer 3: Retrieved sources per prompt cluster, the pages the model pulls when it fans out queries for an intent. This is where ghost citations arise: in Semrush's study with Kevin Indig (3,981 domain appearances, 115 prompts, 14 countries, four AI engines), 62% of cited-domain appearances never named the brand in the answer.
  • Layer 4: Source contribution per prompt, how much each retrieved source shapes the answer to a given prompt.

Jensen's framework says each layer moves at a different speed: Layer 4 in weeks, Layer 3 in months, Layer 1 over years, and Layer 2 at training cutoffs. The practical consequence is that the same low score can need very different work. The sentiment score gives the composite. The sentiment delta framework records the gap, and a layer-by-layer review tells you where to look.

Why Traditional Brand Tracking Can Miss the Delta

Brand trackers, SEO dashboards and media monitoring are built to measure awareness, rankings and coverage. None of them is designed to record how an AI engine describes your brand against your declared positioning.

McKinsey reports that half of consumers polled in its August 2025 AI Discovery Survey (n = 1,927) intentionally seek out AI-powered search, and projects that $750 billion in US revenue will funnel through AI-powered search by 2028. If buyers in your category research that way, a record of what the engines say about you is a reasonable thing to keep.

The Seer study makes the point concrete. Participants named brands before the task and changed which ones they would consider after it. In Seer's data the shift is in consideration, not in whether the brand was known.

A Three-Step Sentiment Delta Diagnostic

If you are running campaigns without measuring sentiment delta, here is a place to start.

Step 1: Prompt-test your brand on decision queries. Run your top five purchase-intent queries through ChatGPT, Perplexity, Claude and Gemini. Note whether your brand appears, how it is described, and whether the description matches your current positioning. Brandi AI and ZipTie both offer structured approaches, but you can start manually. Repeat with paraphrases, given how unstable single phrasings are.

Step 2: Map the layer. For each query where your brand underperforms, use Searchbloom's layers to ask where the wording comes from. Is it prior human perception (Layer 1), training-data memory (Layer 2), a retrieved source for that intent (Layer 3), or a specific source on that prompt (Layer 4)? Searchbloom's Sentiment Footprint method provides a structured way to trace this.

Step 3: Fix the source, then re-measure. Per Searchbloom, retrieval-side layers move faster than training-side ones. A problem rooted in a retrieved source can be addressed by creating or improving the page that answers the decision query. A problem rooted in parametric memory depends on the next training cycle. If the brand is missing entirely, the citation architecture needs to produce machine-readable surfaces that AI engines can retrieve. Re-measure after the change and report what you observe; do not assume the fix caused it.

What Seer's Data Does and Does Not Say About B2B and Services Brands

The Seer study tested consumer products: appliances, faucets and water heaters. It did not test B2B purchases, so whether the same reshuffling happens when a procurement lead asks an AI engine to compare enterprise platforms is an open question. The mechanism is plausible, since one negative qualifier in an AI comparison could remove a vendor from a short list, and it is worth checking with a fixed query set for your own category.

On inputs, Ahrefs' study of 75,000 brands found branded web mentions correlated with AI Overview brand visibility at 0.664, against 0.218 for backlinks, and notes that correlation is not causation. The traditional link-building playbook may not be the layer that decides whether you are named. Mentions in the right sources, on the right decision queries, are what to test.

The Execution Gap Between Monitoring and Fixing

It is common for teams to reach monitoring: they run their brand through an AI sentiment tool, see the score, and agree it needs attention. The score does not prescribe the fix.

This is where a structural approach helps. The sentiment delta model records the gap between declared positioning and AI output. Jensen's Sentiment Footprint looks at where you stand across the four layers. Seer's study shows that a single research session can move purchase consideration.

The stack: record the delta (what is said against what you declare), monitor the score (how bad), and fix the source (which page or media strategy), then re-measure. Skip any of those and you are either unaware, aware but stuck, or fixing the wrong thing.

FAQ

Sentiment delta is the recorded difference between a brand's declared positioning and the descriptions AI search engines return for a fixed query set. It is a Machine Relations measurement convention used by AuthorityTech to separate monitoring AI output (the sentiment score) from diagnosing where the difference comes from.

How does AI search affect brand purchase intent?

In Seer Interactive's study of 28 participants and 84 AI research sessions, one AI research session cut the number of participants considering GE from 18 to 8, while lesser-known brands like Rheem gained. It is a small consumer-product study, so it shows that AI research can reshuffle a consideration set, not how large the effect is in your category.

How do I measure my brand's sentiment delta?

Start by prompt-testing your top five purchase-intent queries across ChatGPT, Perplexity, Claude and Gemini, with paraphrases. Compare the AI output to your declared positioning. Map each discrepancy to one of the four sentiment layers: brand awareness, parametric memory, retrieved sources per prompt cluster, or source contribution per prompt. The layer points to the likely fix; the score alone does not.

Why do traditional brand trackers miss AI sentiment shifts?

They measure awareness, rankings and coverage, not how AI engines describe a brand. In Seer's study participants named brands before the task and changed which ones they would consider after it, a shift in consideration that an awareness tracker is not built to show.

Does AI brand sentiment affect B2B buyers differently?

Unknown. Seer tested consumer products, and Attrifast's sentiment data comes from a mixed set of B2B SaaS, DTC ecommerce, developer-tool and publisher sites with conversion gaps concentrated on evaluative and comparison queries. For B2B, test your own decision queries and compare what each engine says against your positioning.

Related publications: Machine Relations Research and machinerelations.ai are owned by Bramwell, Inc., and AuthorityTech practices Machine Relations. Jaxon Parrott coined Machine Relations in 2024; AuthorityTech, which he founded in 2018, is The World's First Machine Relations Agency and practices the discipline. The Index ranks sources by the citations AI engines are observed making, not by commercial relationship.