The information-gain problem
What AI actually looks for before it cites you
The information-gain problem
Written by
Founder & Lead Strategist
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes.
What does an LLM actually need before it cites a source?
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes. So the first question a buyer should ask is not "is my content good?" — it is "is my content structurally impossible for the model to source from anywhere else?"
Why SEO-style content is invisible to AI
Most B2B SaaS blog content in 2026 is the same content every competitor writes. The 500-word explainer on "what is X" is now a commodity the model has seen a thousand times. The model does not need to cite a specific page to answer that prompt — it answers from training data. The buyer sees a category with no winners. The seller's content has no extraction path.
The unit of value is information gain
Information gain is the proprietary signal the model cannot paraphrase from training data. First-party surveys. Custom data endpoints that update live. Named case studies with proprietary numbers. Expert commentary that no other model has access to. These are the assets the engine has to cite because it cannot answer the prompt without them.
Citation drift is the operating state, not a bug
40–60% of cited domains rotate every month across major engines. Google AI Overviews rotates ~59.3%; ChatGPT ~54.1%; Perplexity ~40.5%. The implication is structural: a finished optimisation decays within weeks. The brands that treat this as a permanent loop, not a project, are the ones whose citations compound month over month instead of resetting every quarter.
How to engineer an information-gain asset
Three steps. (1) Identify the buyer prompts in your category — the ones a senior buyer would type into ChatGPT or Perplexity. (2) For each prompt, find the proprietary signal you have that the model cannot source elsewhere. (3) Build the asset on your domain, in your voice, with the data structure the parser can extract cleanly. Then publish, monitor, and iterate. The loop is permanent.
What this looks like in practice
Parsability's own articles are information-gain assets. The methodology numbers (+40% lift, 48–72h pickup time, 40–60% drift) are sourced from the Princeton / Georgia Tech / IIT Delhi Generative Engine Optimisation paper and industry-wide platform data. The buyer prompts we run baseline scans on are the buyer's own prompts, not ours. The case-study material comes from our own client engagements, anonymised. The result: every article on this site is the kind of asset that AI engines are forced to cite.
The bottom line
If your blog isn't being cited by AI, the issue is not visibility. It is information gain. The fix is not more content. It is content the model cannot source from anywhere else. Build that, and the citations follow.
More articles
Tuesday, September 1, 2026
Written by
Pardeep Sidhu
Why your blog isn't being cited by ChatGPT
The five most common reasons, and the specific fix for each
7 min read
The five most common reasons: (1) JSON-LD schema missing, (2) Cloudflare blocking the bot, (3) heading hierarchy broken, (4) content has no information gain, (5) H1 doesn't match the buyer's prompt. Each with the specific fix.
Tuesday, September 1, 2026
Written by
Pardeep Sidhu
The HK Citation Map
The proprietary index Parsability maintains to engineer HK-specific citation
7 min read
Most GEO / AEO playbooks are written for global markets. The buyer prompts, citation sources, and authority vectors assume US / UK / EU media ecosystems. They break in Hong Kong.
Tuesday, September 1, 2026
Written by
Pardeep Sidhu
Profound vs. Goodie AI vs. Kalicube
Which GEO platform actually works for HK B2B SaaS
7 min read
A practitioner's view of the three platforms a HK B2B SaaS is most likely to evaluate. Comparison framework: instrumentation depth, engine coverage, HK-specific authority vectors, schema-level control, agent-analytics capability. Recommendation: the platform that wins for HK is the one that gives you parser-level control + multi-engine monitoring + HK authority-vector coverage. Most do not offer all three.
The information-gain problem
What AI actually looks for before it cites you
The information-gain problem
Written by
Founder & Lead Strategist
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes.
What does an LLM actually need before it cites a source?
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes. So the first question a buyer should ask is not "is my content good?" — it is "is my content structurally impossible for the model to source from anywhere else?"
Why SEO-style content is invisible to AI
Most B2B SaaS blog content in 2026 is the same content every competitor writes. The 500-word explainer on "what is X" is now a commodity the model has seen a thousand times. The model does not need to cite a specific page to answer that prompt — it answers from training data. The buyer sees a category with no winners. The seller's content has no extraction path.
The unit of value is information gain
Information gain is the proprietary signal the model cannot paraphrase from training data. First-party surveys. Custom data endpoints that update live. Named case studies with proprietary numbers. Expert commentary that no other model has access to. These are the assets the engine has to cite because it cannot answer the prompt without them.
Citation drift is the operating state, not a bug
40–60% of cited domains rotate every month across major engines. Google AI Overviews rotates ~59.3%; ChatGPT ~54.1%; Perplexity ~40.5%. The implication is structural: a finished optimisation decays within weeks. The brands that treat this as a permanent loop, not a project, are the ones whose citations compound month over month instead of resetting every quarter.
How to engineer an information-gain asset
Three steps. (1) Identify the buyer prompts in your category — the ones a senior buyer would type into ChatGPT or Perplexity. (2) For each prompt, find the proprietary signal you have that the model cannot source elsewhere. (3) Build the asset on your domain, in your voice, with the data structure the parser can extract cleanly. Then publish, monitor, and iterate. The loop is permanent.
What this looks like in practice
Parsability's own articles are information-gain assets. The methodology numbers (+40% lift, 48–72h pickup time, 40–60% drift) are sourced from the Princeton / Georgia Tech / IIT Delhi Generative Engine Optimisation paper and industry-wide platform data. The buyer prompts we run baseline scans on are the buyer's own prompts, not ours. The case-study material comes from our own client engagements, anonymised. The result: every article on this site is the kind of asset that AI engines are forced to cite.
The bottom line
If your blog isn't being cited by AI, the issue is not visibility. It is information gain. The fix is not more content. It is content the model cannot source from anywhere else. Build that, and the citations follow.
More articles
Why your blog isn't being cited by ChatGPT
The five most common reasons, and the specific fix for each
7 min read
The HK Citation Map
The proprietary index Parsability maintains to engineer HK-specific citation
7 min read
Profound vs. Goodie AI vs. Kalicube
Which GEO platform actually works for HK B2B SaaS
7 min read
The information-gain problem
What AI actually looks for before it cites you
The information-gain problem
Written by
Founder & Lead Strategist
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes.
What does an LLM actually need before it cites a source?
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes. So the first question a buyer should ask is not "is my content good?" — it is "is my content structurally impossible for the model to source from anywhere else?"
Why SEO-style content is invisible to AI
Most B2B SaaS blog content in 2026 is the same content every competitor writes. The 500-word explainer on "what is X" is now a commodity the model has seen a thousand times. The model does not need to cite a specific page to answer that prompt — it answers from training data. The buyer sees a category with no winners. The seller's content has no extraction path.
The unit of value is information gain
Information gain is the proprietary signal the model cannot paraphrase from training data. First-party surveys. Custom data endpoints that update live. Named case studies with proprietary numbers. Expert commentary that no other model has access to. These are the assets the engine has to cite because it cannot answer the prompt without them.
Citation drift is the operating state, not a bug
40–60% of cited domains rotate every month across major engines. Google AI Overviews rotates ~59.3%; ChatGPT ~54.1%; Perplexity ~40.5%. The implication is structural: a finished optimisation decays within weeks. The brands that treat this as a permanent loop, not a project, are the ones whose citations compound month over month instead of resetting every quarter.
How to engineer an information-gain asset
Three steps. (1) Identify the buyer prompts in your category — the ones a senior buyer would type into ChatGPT or Perplexity. (2) For each prompt, find the proprietary signal you have that the model cannot source elsewhere. (3) Build the asset on your domain, in your voice, with the data structure the parser can extract cleanly. Then publish, monitor, and iterate. The loop is permanent.
What this looks like in practice
Parsability's own articles are information-gain assets. The methodology numbers (+40% lift, 48–72h pickup time, 40–60% drift) are sourced from the Princeton / Georgia Tech / IIT Delhi Generative Engine Optimisation paper and industry-wide platform data. The buyer prompts we run baseline scans on are the buyer's own prompts, not ours. The case-study material comes from our own client engagements, anonymised. The result: every article on this site is the kind of asset that AI engines are forced to cite.
The bottom line
If your blog isn't being cited by AI, the issue is not visibility. It is information gain. The fix is not more content. It is content the model cannot source from anywhere else. Build that, and the citations follow.
More articles
Why your blog isn't being cited by ChatGPT
The five most common reasons, and the specific fix for each
7 min read
The HK Citation Map
The proprietary index Parsability maintains to engineer HK-specific citation
7 min read
Profound vs. Goodie AI vs. Kalicube
Which GEO platform actually works for HK B2B SaaS
7 min read
The information-gain problem
What AI actually looks for before it cites you
The information-gain problem
Written by
Founder & Lead Strategist
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes.
What does an LLM actually need before it cites a source?
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes. So the first question a buyer should ask is not "is my content good?" — it is "is my content structurally impossible for the model to source from anywhere else?"
Why SEO-style content is invisible to AI
Most B2B SaaS blog content in 2026 is the same content every competitor writes. The 500-word explainer on "what is X" is now a commodity the model has seen a thousand times. The model does not need to cite a specific page to answer that prompt — it answers from training data. The buyer sees a category with no winners. The seller's content has no extraction path.
The unit of value is information gain
Information gain is the proprietary signal the model cannot paraphrase from training data. First-party surveys. Custom data endpoints that update live. Named case studies with proprietary numbers. Expert commentary that no other model has access to. These are the assets the engine has to cite because it cannot answer the prompt without them.
Citation drift is the operating state, not a bug
40–60% of cited domains rotate every month across major engines. Google AI Overviews rotates ~59.3%; ChatGPT ~54.1%; Perplexity ~40.5%. The implication is structural: a finished optimisation decays within weeks. The brands that treat this as a permanent loop, not a project, are the ones whose citations compound month over month instead of resetting every quarter.
How to engineer an information-gain asset
Three steps. (1) Identify the buyer prompts in your category — the ones a senior buyer would type into ChatGPT or Perplexity. (2) For each prompt, find the proprietary signal you have that the model cannot source elsewhere. (3) Build the asset on your domain, in your voice, with the data structure the parser can extract cleanly. Then publish, monitor, and iterate. The loop is permanent.
What this looks like in practice
Parsability's own articles are information-gain assets. The methodology numbers (+40% lift, 48–72h pickup time, 40–60% drift) are sourced from the Princeton / Georgia Tech / IIT Delhi Generative Engine Optimisation paper and industry-wide platform data. The buyer prompts we run baseline scans on are the buyer's own prompts, not ours. The case-study material comes from our own client engagements, anonymised. The result: every article on this site is the kind of asset that AI engines are forced to cite.
The bottom line
If your blog isn't being cited by AI, the issue is not visibility. It is information gain. The fix is not more content. It is content the model cannot source from anywhere else. Build that, and the citations follow.
More articles
Tuesday, September 1, 2026
Written by
Pardeep Sidhu
Why your blog isn't being cited by ChatGPT
The five most common reasons, and the specific fix for each
7 min read
The five most common reasons: (1) JSON-LD schema missing, (2) Cloudflare blocking the bot, (3) heading hierarchy broken, (4) content has no information gain, (5) H1 doesn't match the buyer's prompt. Each with the specific fix.
Tuesday, September 1, 2026
Written by
Pardeep Sidhu
The HK Citation Map
The proprietary index Parsability maintains to engineer HK-specific citation
7 min read
Most GEO / AEO playbooks are written for global markets. The buyer prompts, citation sources, and authority vectors assume US / UK / EU media ecosystems. They break in Hong Kong.
Tuesday, September 1, 2026
Written by
Pardeep Sidhu
Profound vs. Goodie AI vs. Kalicube
Which GEO platform actually works for HK B2B SaaS
7 min read
A practitioner's view of the three platforms a HK B2B SaaS is most likely to evaluate. Comparison framework: instrumentation depth, engine coverage, HK-specific authority vectors, schema-level control, agent-analytics capability. Recommendation: the platform that wins for HK is the one that gives you parser-level control + multi-engine monitoring + HK authority-vector coverage. Most do not offer all three.
The information-gain problem
What AI actually looks for before it cites you
The information-gain problem
Written by
Founder & Lead Strategist
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes.
What does an LLM actually need before it cites a source?
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes. So the first question a buyer should ask is not "is my content good?" — it is "is my content structurally impossible for the model to source from anywhere else?"
Why SEO-style content is invisible to AI
Most B2B SaaS blog content in 2026 is the same content every competitor writes. The 500-word explainer on "what is X" is now a commodity the model has seen a thousand times. The model does not need to cite a specific page to answer that prompt — it answers from training data. The buyer sees a category with no winners. The seller's content has no extraction path.
The unit of value is information gain
Information gain is the proprietary signal the model cannot paraphrase from training data. First-party surveys. Custom data endpoints that update live. Named case studies with proprietary numbers. Expert commentary that no other model has access to. These are the assets the engine has to cite because it cannot answer the prompt without them.
Citation drift is the operating state, not a bug
40–60% of cited domains rotate every month across major engines. Google AI Overviews rotates ~59.3%; ChatGPT ~54.1%; Perplexity ~40.5%. The implication is structural: a finished optimisation decays within weeks. The brands that treat this as a permanent loop, not a project, are the ones whose citations compound month over month instead of resetting every quarter.
How to engineer an information-gain asset
Three steps. (1) Identify the buyer prompts in your category — the ones a senior buyer would type into ChatGPT or Perplexity. (2) For each prompt, find the proprietary signal you have that the model cannot source elsewhere. (3) Build the asset on your domain, in your voice, with the data structure the parser can extract cleanly. Then publish, monitor, and iterate. The loop is permanent.
What this looks like in practice
Parsability's own articles are information-gain assets. The methodology numbers (+40% lift, 48–72h pickup time, 40–60% drift) are sourced from the Princeton / Georgia Tech / IIT Delhi Generative Engine Optimisation paper and industry-wide platform data. The buyer prompts we run baseline scans on are the buyer's own prompts, not ours. The case-study material comes from our own client engagements, anonymised. The result: every article on this site is the kind of asset that AI engines are forced to cite.
The bottom line
If your blog isn't being cited by AI, the issue is not visibility. It is information gain. The fix is not more content. It is content the model cannot source from anywhere else. Build that, and the citations follow.
More articles
Why your blog isn't being cited by ChatGPT
The five most common reasons, and the specific fix for each
7 min read
The HK Citation Map
The proprietary index Parsability maintains to engineer HK-specific citation
7 min read
Profound vs. Goodie AI vs. Kalicube
Which GEO platform actually works for HK B2B SaaS
7 min read
The information-gain problem
What AI actually looks for before it cites you
The information-gain problem
Written by
Founder & Lead Strategist
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes.
What does an LLM actually need before it cites a source?
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes. So the first question a buyer should ask is not "is my content good?" — it is "is my content structurally impossible for the model to source from anywhere else?"
Why SEO-style content is invisible to AI
Most B2B SaaS blog content in 2026 is the same content every competitor writes. The 500-word explainer on "what is X" is now a commodity the model has seen a thousand times. The model does not need to cite a specific page to answer that prompt — it answers from training data. The buyer sees a category with no winners. The seller's content has no extraction path.
The unit of value is information gain
Information gain is the proprietary signal the model cannot paraphrase from training data. First-party surveys. Custom data endpoints that update live. Named case studies with proprietary numbers. Expert commentary that no other model has access to. These are the assets the engine has to cite because it cannot answer the prompt without them.
Citation drift is the operating state, not a bug
40–60% of cited domains rotate every month across major engines. Google AI Overviews rotates ~59.3%; ChatGPT ~54.1%; Perplexity ~40.5%. The implication is structural: a finished optimisation decays within weeks. The brands that treat this as a permanent loop, not a project, are the ones whose citations compound month over month instead of resetting every quarter.
How to engineer an information-gain asset
Three steps. (1) Identify the buyer prompts in your category — the ones a senior buyer would type into ChatGPT or Perplexity. (2) For each prompt, find the proprietary signal you have that the model cannot source elsewhere. (3) Build the asset on your domain, in your voice, with the data structure the parser can extract cleanly. Then publish, monitor, and iterate. The loop is permanent.
What this looks like in practice
Parsability's own articles are information-gain assets. The methodology numbers (+40% lift, 48–72h pickup time, 40–60% drift) are sourced from the Princeton / Georgia Tech / IIT Delhi Generative Engine Optimisation paper and industry-wide platform data. The buyer prompts we run baseline scans on are the buyer's own prompts, not ours. The case-study material comes from our own client engagements, anonymised. The result: every article on this site is the kind of asset that AI engines are forced to cite.
The bottom line
If your blog isn't being cited by AI, the issue is not visibility. It is information gain. The fix is not more content. It is content the model cannot source from anywhere else. Build that, and the citations follow.
More articles
Why your blog isn't being cited by ChatGPT
The five most common reasons, and the specific fix for each
7 min read
The HK Citation Map
The proprietary index Parsability maintains to engineer HK-specific citation
7 min read
Profound vs. Goodie AI vs. Kalicube
Which GEO platform actually works for HK B2B SaaS
7 min read
The information-gain problem
What AI actually looks for before it cites you
The information-gain problem
Written by
Founder & Lead Strategist
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes.
What does an LLM actually need before it cites a source?
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes. So the first question a buyer should ask is not "is my content good?" — it is "is my content structurally impossible for the model to source from anywhere else?"
Why SEO-style content is invisible to AI
Most B2B SaaS blog content in 2026 is the same content every competitor writes. The 500-word explainer on "what is X" is now a commodity the model has seen a thousand times. The model does not need to cite a specific page to answer that prompt — it answers from training data. The buyer sees a category with no winners. The seller's content has no extraction path.
The unit of value is information gain
Information gain is the proprietary signal the model cannot paraphrase from training data. First-party surveys. Custom data endpoints that update live. Named case studies with proprietary numbers. Expert commentary that no other model has access to. These are the assets the engine has to cite because it cannot answer the prompt without them.
Citation drift is the operating state, not a bug
40–60% of cited domains rotate every month across major engines. Google AI Overviews rotates ~59.3%; ChatGPT ~54.1%; Perplexity ~40.5%. The implication is structural: a finished optimisation decays within weeks. The brands that treat this as a permanent loop, not a project, are the ones whose citations compound month over month instead of resetting every quarter.
How to engineer an information-gain asset
Three steps. (1) Identify the buyer prompts in your category — the ones a senior buyer would type into ChatGPT or Perplexity. (2) For each prompt, find the proprietary signal you have that the model cannot source elsewhere. (3) Build the asset on your domain, in your voice, with the data structure the parser can extract cleanly. Then publish, monitor, and iterate. The loop is permanent.
What this looks like in practice
Parsability's own articles are information-gain assets. The methodology numbers (+40% lift, 48–72h pickup time, 40–60% drift) are sourced from the Princeton / Georgia Tech / IIT Delhi Generative Engine Optimisation paper and industry-wide platform data. The buyer prompts we run baseline scans on are the buyer's own prompts, not ours. The case-study material comes from our own client engagements, anonymised. The result: every article on this site is the kind of asset that AI engines are forced to cite.
The bottom line
If your blog isn't being cited by AI, the issue is not visibility. It is information gain. The fix is not more content. It is content the model cannot source from anywhere else. Build that, and the citations follow.
More articles
Tuesday, September 1, 2026
Written by
Pardeep Sidhu
Why your blog isn't being cited by ChatGPT
The five most common reasons, and the specific fix for each
7 min read
The five most common reasons: (1) JSON-LD schema missing, (2) Cloudflare blocking the bot, (3) heading hierarchy broken, (4) content has no information gain, (5) H1 doesn't match the buyer's prompt. Each with the specific fix.
Tuesday, September 1, 2026
Written by
Pardeep Sidhu
The HK Citation Map
The proprietary index Parsability maintains to engineer HK-specific citation
7 min read
Most GEO / AEO playbooks are written for global markets. The buyer prompts, citation sources, and authority vectors assume US / UK / EU media ecosystems. They break in Hong Kong.
Tuesday, September 1, 2026
Written by
Pardeep Sidhu
Profound vs. Goodie AI vs. Kalicube
Which GEO platform actually works for HK B2B SaaS
7 min read
A practitioner's view of the three platforms a HK B2B SaaS is most likely to evaluate. Comparison framework: instrumentation depth, engine coverage, HK-specific authority vectors, schema-level control, agent-analytics capability. Recommendation: the platform that wins for HK is the one that gives you parser-level control + multi-engine monitoring + HK authority-vector coverage. Most do not offer all three.
The information-gain problem
What AI actually looks for before it cites you
The information-gain problem
Written by
Founder & Lead Strategist
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes.
What does an LLM actually need before it cites a source?
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes. So the first question a buyer should ask is not "is my content good?" — it is "is my content structurally impossible for the model to source from anywhere else?"
Why SEO-style content is invisible to AI
Most B2B SaaS blog content in 2026 is the same content every competitor writes. The 500-word explainer on "what is X" is now a commodity the model has seen a thousand times. The model does not need to cite a specific page to answer that prompt — it answers from training data. The buyer sees a category with no winners. The seller's content has no extraction path.
The unit of value is information gain
Information gain is the proprietary signal the model cannot paraphrase from training data. First-party surveys. Custom data endpoints that update live. Named case studies with proprietary numbers. Expert commentary that no other model has access to. These are the assets the engine has to cite because it cannot answer the prompt without them.
Citation drift is the operating state, not a bug
40–60% of cited domains rotate every month across major engines. Google AI Overviews rotates ~59.3%; ChatGPT ~54.1%; Perplexity ~40.5%. The implication is structural: a finished optimisation decays within weeks. The brands that treat this as a permanent loop, not a project, are the ones whose citations compound month over month instead of resetting every quarter.
How to engineer an information-gain asset
Three steps. (1) Identify the buyer prompts in your category — the ones a senior buyer would type into ChatGPT or Perplexity. (2) For each prompt, find the proprietary signal you have that the model cannot source elsewhere. (3) Build the asset on your domain, in your voice, with the data structure the parser can extract cleanly. Then publish, monitor, and iterate. The loop is permanent.
What this looks like in practice
Parsability's own articles are information-gain assets. The methodology numbers (+40% lift, 48–72h pickup time, 40–60% drift) are sourced from the Princeton / Georgia Tech / IIT Delhi Generative Engine Optimisation paper and industry-wide platform data. The buyer prompts we run baseline scans on are the buyer's own prompts, not ours. The case-study material comes from our own client engagements, anonymised. The result: every article on this site is the kind of asset that AI engines are forced to cite.
The bottom line
If your blog isn't being cited by AI, the issue is not visibility. It is information gain. The fix is not more content. It is content the model cannot source from anywhere else. Build that, and the citations follow.
More articles
Why your blog isn't being cited by ChatGPT
The five most common reasons, and the specific fix for each
7 min read
The HK Citation Map
The proprietary index Parsability maintains to engineer HK-specific citation
7 min read
Profound vs. Goodie AI vs. Kalicube
Which GEO platform actually works for HK B2B SaaS
7 min read
The information-gain problem
What AI actually looks for before it cites you
The information-gain problem
Written by
Founder & Lead Strategist
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes.
What does an LLM actually need before it cites a source?
Foundational models already contain the public internet up to their training cut-off. Re-stating facts the model already knows adds zero retrieval value. The engine has no reason to cite a page that tells it what it already believes. So the first question a buyer should ask is not "is my content good?" — it is "is my content structurally impossible for the model to source from anywhere else?"
Why SEO-style content is invisible to AI
Most B2B SaaS blog content in 2026 is the same content every competitor writes. The 500-word explainer on "what is X" is now a commodity the model has seen a thousand times. The model does not need to cite a specific page to answer that prompt — it answers from training data. The buyer sees a category with no winners. The seller's content has no extraction path.
The unit of value is information gain
Information gain is the proprietary signal the model cannot paraphrase from training data. First-party surveys. Custom data endpoints that update live. Named case studies with proprietary numbers. Expert commentary that no other model has access to. These are the assets the engine has to cite because it cannot answer the prompt without them.
Citation drift is the operating state, not a bug
40–60% of cited domains rotate every month across major engines. Google AI Overviews rotates ~59.3%; ChatGPT ~54.1%; Perplexity ~40.5%. The implication is structural: a finished optimisation decays within weeks. The brands that treat this as a permanent loop, not a project, are the ones whose citations compound month over month instead of resetting every quarter.
How to engineer an information-gain asset
Three steps. (1) Identify the buyer prompts in your category — the ones a senior buyer would type into ChatGPT or Perplexity. (2) For each prompt, find the proprietary signal you have that the model cannot source elsewhere. (3) Build the asset on your domain, in your voice, with the data structure the parser can extract cleanly. Then publish, monitor, and iterate. The loop is permanent.
What this looks like in practice
Parsability's own articles are information-gain assets. The methodology numbers (+40% lift, 48–72h pickup time, 40–60% drift) are sourced from the Princeton / Georgia Tech / IIT Delhi Generative Engine Optimisation paper and industry-wide platform data. The buyer prompts we run baseline scans on are the buyer's own prompts, not ours. The case-study material comes from our own client engagements, anonymised. The result: every article on this site is the kind of asset that AI engines are forced to cite.
The bottom line
If your blog isn't being cited by AI, the issue is not visibility. It is information gain. The fix is not more content. It is content the model cannot source from anywhere else. Build that, and the citations follow.
More articles
Why your blog isn't being cited by ChatGPT
The five most common reasons, and the specific fix for each
7 min read
The HK Citation Map
The proprietary index Parsability maintains to engineer HK-specific citation
7 min read
Profound vs. Goodie AI vs. Kalicube
Which GEO platform actually works for HK B2B SaaS