Why Am I Not Appearing in ChatGPT Recommendations?

Most businesses diagnose their AI visibility problem as an authority problem.

They assume the model does not know them, does not trust them, or has not indexed enough about them. So they produce more content. They pursue more backlinks. They wait.

That is not the problem. Or rather, it is not the first problem.

The answerability filter

Before any authority signal is evaluated, the model runs a different check entirely. It asks: can this content be extracted and reformatted as a direct answer? If the answer is no, the business does not progress further. It is filtered out before reputation, trust, or scale is ever considered.

This is the answerability filter. It is not theory. It shows up every time we conduct an AI visibility assessment, regardless of business size, domain authority, or content volume.

How AI evaluates businesses

Human searchers scan for relevance and click through. AI models do not. When a query is submitted, the model does not retrieve pages and rank them. It compresses candidate information into a synthesised response. To do that, it needs content that is already structured as an answer.

The model evaluates whether a passage can be lifted, paraphrased, and delivered confidently without ambiguity. This happens before authority, trust, or reputation. Businesses that fail this check are not penalised. They are simply not considered.

Structured answerability: the core mechanism

Structured answerability is the degree to which a piece of content contains a complete, self-contained response to a probable query, expressed in language the model can compress without inference.

It is not keyword density. It is not length. It is internal coherence: a claim, its supporting context, and a resolution, all within a retrievable passage (Liu et al., 2023).

The model evaluates this through pattern recognition trained on Q&A corpora, encyclopaedic entries, and documentation (Brown et al., 2020). Content that mirrors those structural patterns passes. Content that does not, fails silently.

Common failures

Capability framing without resolution.

Pages describe what a business does, not what outcome a client receives or how. The model cannot extract a confident answer from a capability list (Metzler et al., 2021).

Buried specificity.

The most answerable sentence sits in the fourth paragraph, after context-setting the model cannot use. One of the first checks in our assessments is whether the key claim appears within the first 80 words of a passage.

Implicit expertise.

The business assumes the reader understands the domain. The model does not assume. Unqualified claims without brief definitional support fail the compression test (Petroni et al., 2021).

Inconsistent entity description.

When a business describes itself differently across pages, the model cannot form a stable entity profile. Instability signals low reliability (Yao et al., 2023).

Patterns in success

When we analyse brands that appear consistently in AI-generated answers, four commonalities emerge.

• Their core service pages lead with a declarative statement: what they do, for whom, and with what outcome.
• They define specialist terms briefly before using them.
• They repeat consistent entity descriptors, the same name, category, and geographic scope, across all surfaces.
• They answer likely follow-up questions within the same passage, reducing the model's need to infer.

What we do not see are shortcuts. Not high domain authority compensating for structural ambiguity. Not content volume substituting for passage-level coherence.

The counter-intuitive finding

Clarity outperforms scale. A single, well-structured service page from a small business will outperform ten loosely written pages from an established brand. The model does not reward effort or presence. It rewards extractability (Izacard and Grave, 2021).

This directly challenges the assumption that AI visibility is an authority problem solvable through content production.

Practical requirements

Audit your primary service pages using these diagnostic questions:

• Does the opening paragraph answer who you serve, what you deliver, and what changes as a result?
• Is every specialist term defined within the same passage where it is used?
• Is your business described using identical categorical language across your website, directory listings, and third-party mentions?

Specific requirements include:

• A declarative lead sentence on every core page.
• Entity descriptors that match across all indexed surfaces.
• Removal of preamble that delays the answerable claim beyond 80 words.
• Explicit outcome statements rather than process descriptions.

References

• Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P. and Amodei, D. (2020) 'Language models are few-shot learners', Advances in Neural Information Processing Systems, 33, pp. 1877–1901.
• Izacard, G. and Grave, E. (2021) 'Leveraging passage retrieval with generative models for open domain question answering', Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pp. 874–880.
• Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N. and Kiela, D. (2020) 'Retrieval-augmented generation for knowledge-intensive NLP tasks', Advances in Neural Information Processing Systems, 33, pp. 9459–9474.
• Liu, N.F., Lin, K., Hewitt, J., Paranjape, A., Bevilacqua, M., Petroni, F. and Liang, P. (2023) 'Lost in the middle: How language models use long contexts', Transactions of the Association for Computational Linguistics, 12, pp. 157–173.
• Metzler, D., Yi, T., Tay, Y. and Nalisnick, E. (2021) 'Rethinking search: Making domain experts out of dilettantes', ACM SIGIR Forum, 55(1), pp. 1–27.
• OpenAI (2023) GPT-4 technical report. Available at: https://openai.com/research/gpt-4 (Accessed: 10 February 2026).
• Petroni, F., Lewis, P., Piktus, A., Rocktäschel, T., Wu, Y., Miller, A.H. and Riedel, S. (2021) 'How context affects language models' factual predictions', Proceedings of the AKBC Conference. Available at: https://arxiv.org/abs/2005.04611 (Accessed: 10 February 2026).
• Robertson, S. and Zaragoza, H. (2022) 'The probabilistic relevance framework: BM25 and beyond', Foundations and Trends in Information Retrieval, 3(4), pp. 333–389.
• Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K. and Cao, Y. (2023) 'ReAct: Synergising reasoning and acting in language models', International Conference on Learning Representations. Available at: https://arxiv.org/abs/2210.03629 (Accessed: 10 February 2026).
• Zhu, Y., Wichers, N., Lin, C., Wang, X., Chen, T., Shu, L. and Meng, L. (2023) 'Solving math word problems via cooperative reasoning induced language models', Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, pp. 4471–4485.