Being chosen over being seen.
Reach has value only when it ends in a choice. Being on the shortlist is the precondition for being chosen at all3.
We exist to help brands be chosen.
Signed, Souleiman Soulami
Based on the structured literature review Generative AI and the Consumer Choice Process.
Version 1.0
Public
Public
SOULX AGENCY
19.10.2026
THE MANIFESTO
OF
ON BEING CHOSEN
A declaration of how customers choose, where that choice is heading, and how we work because of it, with the articles and evidence behind it
Issued bySouleiman SoulamiFounder and Managing Director
PART I
The Manifesto
PREAMBLE
Whereas customers increasingly ask an AI assistant before they ever visit a website19, and click through to a website roughly half as often once an AI has answered18;
Whereas customers have always chosen from a small shortlist9, and AI increasingly compiles that shortlist on their behalf3,7;
Whereas this shift has only begun, and AI is moving from guiding the search towards making the choice3,17;
We therefore declare the following.
DECLARATIONS
What others say over what you say.
AI draws mostly on reviews, retailers and forums, not on brand websites16. Proof from others outweighs a famous name6.
One sharp need over broad awareness.
Brands are found on the question that describes a specific need, not on general fame15.
Truth over manipulation.
AI can be misled, and customers believe what it tells them11,21. We only publish what is true.
Closing over explaining.
Fewer visitors click through, and those who do arrive better informed1,18. A website must make saying yes easy.
Demand over clicks.
Paid media earns its keep by creating interest in a brand, the signal most closely tied to AI recommendations15, not by buying clicks alone4.
Measured over assumed.
Few brands know whether AI names them16, and poor tracking hides what really works10. We measure before we decide.
Routine over campaigns.
A brand is met at many places it does not control13. Proof and consistency must be a habit, not a one-off.
DECLARATION
Part II sets out the principles, practices and evidence behind each declaration.
PART II
The Articles
The principles behind each declaration, how we apply them, and the evidence they rest on.
CHAPTER I
Foundations
Being chosen over being seen.
Reach has value only when it ends in a choice. We measure our work by whether customers choose a brand, not by how many saw it.
In practiceEvery plan starts from the moment of choice and works backwards: what must a customer see, read and believe to pick this brand?
BasisAI shopping agents concentrate their choices on a handful of products and almost ignore the rest3, and decision aids shrink the shortlist a customer considers8.
Evidence · Moderate
What others say over what you say.
Reviews, customers and independent mentions increasingly decide whether a brand is named. A slogan does not.
In practiceWe map where a brand is discussed outside its own channels and make those places part of the strategy.
BasisBrand websites supply only about five to ten percent of the sources that AI models draw on16. In controlled tests, a well-known brand lost its advantage to a competitor rated just over a tenth of a star higher6.
Evidence · Moderate
One sharp need over broad awareness.
A brand that states precisely whom it serves and which problem it solves is found on the question that matters.
In practiceWe define each brand by the specific situations its customers describe, and build messages and pages around those.
BasisLarge brands that AI assistants omitted for a general category question were recommended once the question described a specific need15.
Evidence · Limited
Systems over tricks.
What works in an algorithm today may not work tomorrow. We build on what lasts: a clear proposition, real proof and a sound funnel.
In practiceWe do not sell hacks. Every tactic we use must still make sense if the platform changes its rules.
BasisContent optimised for generative engines can raise its visibility by up to 40 percent2, yet that advantage largely disappears once competitors apply the same technique6, and a single model update can shift market shares3.
Evidence · Moderate
Truth over manipulation.
We never invent claims, reviews or evidence to influence what an AI says about a brand. What we publish must be true for people and for machines.
In practiceEvery claim we write can be traced to the client's own material or to a verifiable source.
BasisLanguage models can be steered by hidden text11 and by authoritative-sounding claims, including fabricated clinical evidence6. Customers tend to adopt what an AI tells them, even when it is wrong21.
Evidence · Moderate
Owner to owner over agency to client.
Clients deal with the people who do the work. We speak about revenue and customers, in their language.
In practiceNo account managers in between. The people who build and run the work are the people at the table.
BasisPeople rely on algorithms for objective tasks but prefer human judgement for decisions they see as personal or subjective5,14. This is not yet tested for generative AI; choosing a growth partner is, in our reading, such a decision.
Evidence · Limited
CHAPTER II
Websites that are found and chosen
Closing over explaining.
Visitors increasingly arrive already informed. A website's first task is to make saying yes easy.
In practiceShort routes to contact or purchase, fast pages, and the decision information on the first screen.
BasisAI summaries reduce clicks to websites18, while visitors referred by AI stayed longer and converted more often than other visitors1.
Evidence · Limited
The customer's question over our keywords.
Customers no longer type two keywords. They describe their situation. A website must answer that situation, in the customer's own words.
In practicePages are written around real customer questions, collected from sales calls, reviews and search data.
BasisPeople using AI search ask fewer but more complex questions21, and brands absent from general answers appear once the question describes a specific need15.
Evidence · Limited
Structure over decoration.
Every website we build states its facts plainly and in structure: what is offered, for whom, at what price, with what proof. A machine that reads the page must understand it as well as a person does.
In practiceClear headings, structured product and service data, and plain text that does not depend on visuals to be understood.
BasisAI agents also choose from text-only versions of a shop, and product descriptions adapted to the question measurably increased market share3.
Evidence · Limited
Proof at the decision over proof on a separate page.
Ratings, reviews and results belong where the customer decides.
In practiceReviews and results sit next to every call to action, not on a testimonials page nobody visits.
BasisSmall differences in ratings outweighed brand familiarity in AI recommendations6, and the perceived quality and credibility of reviews shape purchase decisions12.
Evidence · Moderate
CHAPTER III
Paid visibility
Paid access over shrinking reach.
As AI answers absorb more searches, fewer customers click through by themselves. Paid visibility keeps the door open where organic traffic shrinks.
In practiceSearch and social campaigns that capture the customers who still click, aimed at the pages that close.
BasisAbout 60 percent of searches end without a click4, and an AI summary halves the share of users who click a result18.
Evidence · Moderate
Demand over clicks.
We use paid media to make people search for and talk about a brand. That interest is what later makes the brand appear in AI answers.
In practiceCampaigns are judged on conversions and on the brand search they create, not on clicks alone.
BasisOf all marketplace signals tested, search interest showed the strongest association with how prominently AI recommended a brand; ad spend on its own added little15. The relationship is correlational.
Evidence · Limited
Earned proof over sponsored placement.
Paid placement alone does not win the recommendation. We combine paid visibility with proof that others confirm.
In practiceEvery paid campaign is paired with the reviews and results that make the click worth trusting.
BasisAI shopping agents consistently penalised products labelled as sponsored and rewarded platform endorsements3.
Evidence · Limited
Where attention goes over where it was.
We bring brands to new channels as customers move there, including advertising inside AI assistants.
In practiceWe test new channels early with small budgets and scale what proves itself.
BasisAttention moves quickly between AI assistants: within one year the leading assistant's share of AI traffic fell from about three quarters to about half, while others grew several times over19.
Evidence · Moderate
CHAPTER IV
Data and tracking
Measured choice over counted clicks.
The goal is to know whether and how a brand appears in AI answers: how often it is named, and how high.
In practiceThis is the direction we are building towards. Until it is in place, we measure what precedes it: brand search, reviews and traffic referred by AI.
BasisFewer than one in five brands track their presence in AI answers16. Researchers measure it as prevalence, how often a brand is named, and prominence, how high it ranks15.
Evidence · Limited
Repeated measurement over single checks.
One check proves nothing. We measure on a fixed rhythm.
In practiceFixed reporting cycles with every figure shown against the previous period.
BasisAI answers vary between runs, so reliable measurement needs repeated sampling15, and model updates can shift market shares substantially3.
Evidence · Moderate
Each source on its own terms over one blended average.
Visitors from AI behave differently from visitors from search or social. We measure and optimise each source separately.
In practiceDashboards split results by source, so a strong channel is never hidden by a weak one.
BasisAI-referred visitors stayed longer, viewed more pages and converted more often than other visitors1.
Evidence · Limited
Sound measurement over fast optimisation.
Bad tracking hides real effects. Before we optimise anything, we make sure the measurement is right.
In practiceEvery engagement starts with a tracking audit; nothing is optimised on data we have not verified.
BasisThe effect of AI recommendations on shop visits looked close to zero with coarse measurement and became visible only with more precise tracking10.
Evidence · Limited
CHAPTER V
Processes that scale being chosen
Proof as routine over proof as campaign.
Collecting reviews, results and testimonials is a weekly routine, built into how a company serves its customers.
In practiceWe help clients set up a fixed moment and method to ask every satisfied customer for proof.
BasisAI draws mostly on third-party sources16 and weighs ratings heavily6. This article is derived from those findings, not tested directly.
Evidence · Limited
One truth over many versions.
Prices, services, opening hours and claims must be identical on the website, in profiles, with partners and in reviews.
In practiceOne source document with the brand's facts, checked against every channel on a fixed rhythm.
BasisCustomers meet a brand at touchpoints it owns and at many it does not13, and they tend to accept what an AI tells them, even when it is wrong21.
Evidence · Moderate
Rhythm over reaction.
Because the ground keeps moving, we review and adjust in cycles: measure, decide, change, measure again.
In practiceWeekly operational reviews and a periodic strategic review with the client.
BasisMarket shares inside AI shift with each model update3, and any technical advantage fades once competitors catch up6.
Evidence · Limited
CHAPTER VI
Accountability and limits
Outcomes over reports.
We are judged by results a client can see in revenue and customers, not by the length of our reports.
In practiceEvery report ends with what we did, what we saw, what we do next, and what needs a decision.
BasisThis article is a commitment, not a research finding.
Commitment
Evidence over certainty.
We say what we know, what we expect, and what is not yet proven. This charter will change as the evidence does.
In practiceEach article carries its evidence level, and this document is revised when new research appears.
BasisThe evidence base is young: much of it consists of preprints and industry data, mostly from the United States and about products rather than services, and no study yet measures causally whether an AI recommendation leads to a purchase20.
Evidence · Stated limits
THE FOUR DISCIPLINES
Accordingly, SOULX helps companies to be chosen through four disciplines.
- Websitesthat can be found by people and machines, and that close the choice. Chapter II.
- Paid visibilityon search, social and AI assistants, to drive traffic, demand and conversion. Chapter III.
- Data and trackingthat measure what matters and steer every optimisation. Chapter IV.
- Internal processesthat turn proof, consistency and adaptation into routine, so being chosen scales. Chapter V.
REFERENCES
1Adobe Digital Insights. (2026). AI-driven traffic surges across industries. Adobe. https://business.adobe.com/blog/ai-driven-traffic-surges-across-industries
2Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., & Deshpande, A. (2024). GEO: Generative engine optimization. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. https://arxiv.org/abs/2311.09735
3Allouah, A., Besbes, O., Figueroa, J. D., Kanoria, Y., & Kumar, A. (2025). What is your AI agent buying? [Preprint]. arXiv. https://arxiv.org/abs/2508.02630
4Bain & Company. (2025). Goodbye clicks, hello AI: Zero-click search redefines marketing. https://www.bain.com/insights/goodbye-clicks-hello-ai-zero-click-search-redefines-marketing/
5Castelo, N., Bos, M. W., & Lehmann, D. R. (2019). Task-dependent algorithm aversion. Journal of Marketing Research, 56(5), 809–825.
6Chu, X., & Hou, Y. (2026). Incumbent advantage: Brand bias and cognitive manipulation dynamics in LLM recommendation systems [Preprint]. arXiv. https://arxiv.org/abs/2606.17443
7Dellaert, B. G. C., Shu, S. B., Arentze, T. A., Baker, T., Diehl, K., Donkers, B., Fast, N. J., Häubl, G., Johnson, H., Karmarkar, U. R., Oppewal, H., Schmitt, B. H., Schroeder, J., Spiller, S. A., & Steffel, M. (2020). Consumer decisions with artificially intelligent voice assistants. Marketing Letters, 31(4), 335–347. https://doi.org/10.1007/s11002-020-09537-5
8Häubl, G., & Trifts, V. (2000). Consumer decision making in online shopping environments: The effects of interactive decision aids. Marketing Science, 19(1), 4–21.
9Hauser, J. R., & Wernerfelt, B. (1990). An evaluation cost model of consideration sets. Journal of Consumer Research, 16(4), 393–408.
10Iannelli, M., & Ai, A. (2026). From prompt to purchase: How AI brand recommendations move consumers on the open web [Preprint]. arXiv. https://arxiv.org/abs/2606.10907
11Kumar, A., & Lakkaraju, H. (2024). Manipulating large language models to increase product visibility [Preprint]. arXiv. https://arxiv.org/abs/2404.07981
12Lei, K., & Liu, Y. (2025). When AI becomes a shopping advisor: A study on the impact of generative AI review on consumer purchase decision. SAGE Open. https://doi.org/10.1177/21582440251357671
13Lemon, K. N., & Verhoef, P. C. (2016). Understanding customer experience throughout the customer journey. Journal of Marketing, 80(6), 69–96.
14Longoni, C., Bonezzi, A., & Morewedge, C. K. (2019). Resistance to medical artificial intelligence. Journal of Consumer Research, 46(4), 629–650.
15Malthouse, E., Lee, K.-Y., Yang, J., Pal, S., & Feng, X. (2026). Evaluating brand retrieval and ranking in large language model recommendations [Preprint]. arXiv. https://arxiv.org/abs/2609.16304
16McKinsey & Company. (2025). New front door to the internet: Winning in the age of AI search. McKinsey & Company.
17NMI. (2026, August 26). Consumers want AI to help them shop, not control their spending. https://www.nmi.com/about-us/news/nmi-research-consumers-want-ai-to-help-them-shop-not-control-their-spending/
18Pew Research Center. (2025, July 22). Google users are less likely to click on links when an AI summary appears in the results. https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears/
19Similarweb. (2026). Generative AI website traffic share [Data set]. Reported by PPC Land. https://ppc.land/chatgpt-drops-to-52-7-as-claude-triples-its-ai-traffic-share/
20SOULX. (2026). Generative AI and the consumer choice process: A structured literature review [Unpublished internal report].
21Spatharioti, S. E., Rothschild, D. M., Goldstein, D. G., & Hofman, J. M. (2023). Comparing traditional and LLM-based search for consumer choice: A randomized experiment [Preprint]. arXiv. https://arxiv.org/abs/2307.03744
22StatCounter. (2026). AI chatbot market share, Europe and North America, September 2026 [Data set]. https://gs.statcounter.com/ai-chatbot-market-share/all/europe