AEO 101Single source of truth on AEO

For PR and communications teams

Be the narrative AI tells when asked about your brand.

Every prompt about your company already has an answer in ChatGPT, Claude, and Perplexity. We engineer the source pool that decides what that answer says.

By Subia Peerzada/Updated May 14, 2026

§01 How does the new buying funnel actually work?

A reporter, an investor, or a buyer asks AI about your brand. AI assembles a two-sentence reputation.

01The buyer prompt
>_
Reporter, tell me about [your brand], are they credible?
02Retrieval fanout
wikiwikipedia.org
tier-1 presswsj.com / ft.com
tradetrade publication
linkedinexecutive bios
ownedyourbrand.com
03The named answer
[Your brand] is a [category] company known for…
Credibility signals: WSJ feature 2026, Fortune 500
Risks cited: lawsuit in 2024 (resolved)
Leadership: CEO Subia Peerzada, formerly…

The two-sentence summary AI gives is your reputation now. Engineering it is the comms job.

That summary repeats verbatim across surfaces. If a hostile narrative wins the source pool, AI repeats it on every read.

§02 What happened to the old buying funnel?

The 20-minute reputation scan collapsed to one AI prompt.

Pre-AI funnel2023
  1. Reporter or buyer Googles the brand
  2. Reads the top 3 stories
  3. Checks Wikipedia and LinkedIn
  4. Skims a trade publication
  5. Forms an opinion over 20 minutes

5 steps

AI-answer funnel2026
  1. Asks ChatGPT or Perplexity about the brand
  2. Reads AI's named-source summary
  3. Forms an opinion in 90 seconds

3 steps

First-impression formation collapsed to one AI turn. The opinion is now set before your comms team gets a chance to brief.

Comms work shifted from placing the story to engineering the source pool the story gets retrieved from.

§03 Which sources does AI actually read from?

AI's view of your brand comes from a knowable set of surfaces. Most comms teams underweight half of them.

The source pool AI reads from

What we influence, tier by tier

01 · Tier 1Wikipedia and Wikidatawikipedia.org · wikidata.orgAmong the most heavily weighted retrieval and training sources across major LLMs. Stale or inaccurate entries show up verbatim in AI answers.
02 · Tier 1Major editorial presswsj.com · ft.com · nytimes.com · the economistNews-desk stories carry far higher weight than contributor or council posts. Composition of recent coverage decides the sentiment frame.
03 · Tier 2Trade publicationscategory trade press · vertical newsletters · analyst notesNarrative authority inside the industry. Trade coverage shapes how AI describes the brand inside category prompts.
04 · Tier 2Executive bio surfaceslinkedin.com · about pages · conference speaker biosEntity graph signal. Where AI looks to verify the leadership story it tells about the company.
05 · Tier 3Owned and earned long tailnewsroom · podcast appearances · bylined piecesCompounding asset. Lower individual weight, high collective weight when refreshed regularly.

These five tiers carry your reputation. The work is making them agree on the same story.

When the surfaces disagree, AI picks the version with the strongest authority signal. The job is to make them agree.

§04 What metric actually decides the category?

Sentiment composition on the brand prompt set, broken out by surface.

Citation share visualisation

Prompt: is [your brand] a credible company in [category]?

Positive sentimentMixed / hostileLong tail
ChatGPT58% · 28% · 14%
Positive
Mixed
Claude51% · 32% · 17%
Positive
Mixed
Perplexity46% · 35% · 19%
Positive
Mixed
Gemini44% · 38% · 18%
Positive
Mixed
AI Overviews49% · 33% · 18%
Positive
Mixed

Illustrative sentiment composition. Real engagements run a brand prompt set of 40 to 100 prompts weekly with sentiment delta and source attribution.

Each platform tells a slightly different story. Moving the bars means moving the source pool, one cited surface at a time.

§05 What do we actually ship?

Six lines of work, run weekly, owned by us.

Each block describes the actual work, not a tool we hand over. We carry production, the publication relationships, and the platform monitoring.

01

Brand sentiment audit across the five AI surfaces

We run a fixed set of brand and category prompts against ChatGPT, Claude, Gemini, Perplexity, and AI Overviews. The output is a documented read of what AI says about your brand today, where the sentiment came from, and which cited sources are doing the work.

02

Citation attribution from press placements

When a Forbes, WSJ, or trade-press placement runs, the comms team needs to know whether it entered the AI source pool. We trace citation lift on the relevant prompts inside seven days of the placement, and we report which placements moved AI and which did not.

03

Target publication selection by source-pool position

A small set of publications dominate the cited source pool for each category. We tell you which outlets AI actually reads from for your topics, so the comms calendar is built around placements that compound visibility instead of vanity hits.

04

Narrative consistency across surfaces

AI answers drift when your owned site, Wikipedia, LinkedIn, executive bios, and trade-press coverage describe the brand differently. We reconcile the narrative across surfaces and monitor for drift weekly.

05

AI-citation-grade press strategy

Most press releases are not written for AI extraction. We work with the comms team on release structures, quote formats, and supporting data assets that get cited at a higher rate.

06

Weekly sentiment and citation monitoring

A crisis hits, a competitor places a hostile story, an analyst note lands. We monitor the brand prompt set weekly and flag movement before it becomes the consensus answer. Response window measured in days, not quarters.

§06 The methodology is public

One framework, applied weekly. Research, playbook, and engineering ledger all open.

The methodology

CITE framework

Comprehend, Influence, Track, Evolve

The living playbook

AEO 101

Refreshed daily from our research library

The research

AI Visibility Index

Open library of operator-grade briefs

§07 Questions buyers ask before they engage

The questions comms leaders ask before they engage.

How do AI systems decide which press to cite?
Models do not cite press uniformly. Each LLM and answer engine pulls from a recurring set of trusted domains for a given topic, and that set is narrower than most comms teams assume. The factors that decide inclusion are domain authority, topic relevance, recency, structural readability (clear headings, quoted attributions, clean passage structure), and the presence of supporting data or named expert sources inside the piece. A Forbes contributor post and a WSJ news desk story are not weighted equally even though both appear on prestigious mastheads. Knowing which outlets AI actually reads for your category is half the battle.
Why does my Forbes mention not show up in ChatGPT?
Three common reasons. First, the piece is a Forbes Council or contributor post on a subdomain that AI weighs lower than the main editorial line. Second, the piece is well written but does not contain the structured passages AI extracts: clean attributions, quoted expertise, supporting statistics. Third, the brand mention is in passing rather than in a position that signals the article is about the brand. Earned media has to be engineered for AI extraction, not just for the placement itself.
Can negative sentiment in AI answers be fixed?
Yes, but the path is slower than fixing a Google ranking. AI answers reflect a weighted average of the source pool, so the fix is to change the composition and recency of that pool. That means new authoritative coverage, refreshed owned-site content, corrected Wikipedia where applicable, and direct outreach to the publications that carry the negative framing. Most sentiment corrections we run take three to six months of consistent source-pool work to land. The pilot model is well suited to it because the goal is specific and measurable.
Does Wikipedia and Wikidata still matter?
Yes, more than most comms teams assume. Wikipedia is one of the most heavily weighted training and retrieval sources across major LLMs, and Wikidata feeds entity graphs that decide whether a brand has a clean machine-readable identity. Inaccurate or stale Wikipedia content shows up directly inside AI answers, sometimes word for word. Wikipedia is also one of the harder surfaces to touch ethically; we work with experienced editors and follow the platform's notability and conflict-of-interest rules.
How do I attribute AI visibility back to specific PR placements?
Through a controlled prompt set and a citation log. We instrument your category and brand prompts on day one. Every placement is tagged. When a new piece runs, we re-run the prompts inside the same week and document any change in cited sources or sentiment. Over a quarter, the comms team gets a clear read of which placements entered the source pool, which did not, and how each one moved the metrics that matter.

More vertical playbooks from Cite Solutions

Ecommerce & DTC→Consumer Apps→Professional Services→Automotive→Travel→Dev Tools→

Ready to become the answer AI gives?

Book a 30-minute discovery call. We'll show you what AI says about your brand today. No pitch. Just data.

Book a Discovery Call
.md