Somewhere in Harlow & Pike’s inbox sits an AI search optimization proposal. It lists eight line items, each phrased like a gap the firm has been leaving open. The firm is made up, a regional accounting practice with 40 employees, but it stands in for any owner holding a pitch like this one. Its managing partner has no search background and one fair question. Is this new spend, or work the firm already funds?
Our reading of the evidence is that AI search visibility follows recognition, meaning how widely other people talk about a business. Google treats its AI features as ordinary SEO, and OpenAI describes ChatGPT search access as a crawler setting. A separate package therefore has to earn its place line by line. What’s left after that sorting is mostly a coordination job across content, referrals, press, and reputation.
Where Do AI Answers Get Their Sources?
A September 2025 preprint from the University of Toronto found that AI search engines cite earned media far more often than brand-owned websites. Earned media is the coverage, reviews, and mentions that outside parties write. The authors tested several AI engines against Google across industries, languages, and rephrased queries. They also report a big-brand bias that favors established names.
Since it’s a single preprint built on 2025 engines and consumer-style prompts, read it as one study of which kinds of sources get cited. Our jump from there to recognition is an inference, and the label is ours. Still, the direction matters for Harlow & Pike. It points attention toward what trade publications, the local press, and happy clients say about the firm.
What Do Google and OpenAI Say a Site Needs?
Google’s guide, updated in July 2026, says the same SEO fundamentals apply to its AI features. That means site owners can skip llms.txt files, content chunking, and rewriting pages just for AI systems. A page still has to be indexed and eligible to appear with a snippet. OpenAI’s publisher FAQ is narrower, since it treats ChatGPT search visibility as a crawler question. Its advice is to check that robots.txt isn’t blocking OAI-SearchBot if you want content included in ChatGPT summaries and snippets.
Scope limits. Google’s guide covers Google’s own AI features, such as AI Overviews and AI Mode. OpenAI’s FAQ covers ChatGPT search only. Neither document speaks for other AI engines.
If the proposal charges for crawler access and indexing checks, those belong to ordinary website upkeep. The firm’s current web or SEO vendor may already be covering them.
Are Statistics and Citations a Separate Service?
A 2024 KDD study found that adding citations, quotations, and statistics improved a page’s visibility in generative engine responses by up to 40%. Keyword stuffing, meanwhile, gave little to no improvement.
The tests had limits, mostly because the tactics were applied to pages already retrieved from Google’s top five results. The findings cover how a retrieved page reads, while how a page gets retrieved sits outside them. The Perplexity test used uploaded source text on a 200-query subset. Its best gains were 22% or 37%, depending on the tactic and metric.
Those habits belong to editing. A partner quote, a named government source, and one verified figure are what the study tested, and any capable writer can add them. A guide to estimated tax deadlines that names the IRS page and quotes a partner already qualifies, so there’s no new service to buy.
Can a Smaller Business Earn Mentions AI Engines Notice?
It can, though it starts behind, since the Toronto preprint reports a big-brand bias in what AI engines cite. A study from Ahrefs, which sells an AI visibility tracker, points the same direction. It analyzed 75,000 brands with a Domain Rating above 40. Branded web mentions correlated at 0.664 with AI Overview visibility, against 0.218 for backlinks. Ahrefs itself notes that correlation isn’t causation. A Domain Rating filter also tilts the sample toward established domains, so the figures work better as a hint than as a rule.
Google adds a warning about bought mentions. Its guide says chasing inauthentic mentions pays off less than it appears, because ranking systems focus on high-quality content and separate systems block spam. Any package that sells placements deserves the hardest look.
Authentic recognition is slower, but it can start with things a firm like this already controls:
- A partner quoted in a regional business journal
- A speaking slot at a state CPA society event
- Client stories published with permission
None of that guarantees an AI engine will cite the firm, and larger rivals will keep their head start. We treat branded search growth as one recognition signal. That’s why our guide to SEO services that end reactive marketing lists it among the metrics a proactive program tracks.
How Do You Sort an AI Search Optimization Proposal?
Sort each line by what the sources say about it, then mark it already covered, ask for evidence, or fair new work. Google states that third-party tools and services can’t guarantee performance, and it points owners to Search Console for first-party data. The Generative AI performance report there shows impressions from AI Overviews and AI Mode by page, country, device, and date. OpenAI adds that ChatGPT referral links carry a utm_source=chatgpt.com tag, so analytics platforms can track them. The replacement rule in our 2027 marketing plan guide applies too. A new line item earns a place only when something with less evidence behind it leaves.
What Does Each Line Item Cover?
Here’s how the usual proposal lines sort against the sources above:
| Proposal line | What the sources say | Verdict |
|---|---|---|
| llms.txt and special files | Google says site owners don’t need AI text files or new machine-readable files. Keeping an llms.txt file for other services neither helps nor harms visibility in Google Search. | Ask for evidence |
| Content chunking | Google says no requirement exists to break pages into small pieces for AI. | Ask for evidence |
| Rewriting pages for AI | Google says pages don’t need to be written in a specific way for generative AI search. | Ask for evidence |
| Standard schema markup | Google says structured data still supports rich results, so a regular SEO plan may already include it. | Already covered |
| AI-only schema package | Google says structured data isn’t required for generative AI search. | Ask for evidence |
| Bought mentions | Google says chasing inauthentic mentions pays off less than it appears. | Ask for evidence |
| Statistics, quotes, citations | The 2024 KDD study found gains on retrieved pages, within the limits above. | Already covered |
| AI visibility reporting | Google’s report and analytics referral tags cover part of the picture, so extra reporting is fair when it adds something they don’t. | Fair new work |
A line that adds real writing, outreach, or measurement beyond these rows counts as fair new work. It should arrive with its own scope and review date.
What Should You Ask Before Signing?
Agencies like ours send pitches too, and Silesky sells blogging and SEO. Check the table against the linked sources before you rely on our reading. Four questions apply to any line in the proposal:
- Which AI engine does this line target, and which published guidance supports it?
- What would it change on a page or off the site, and could our current SEO work already cover it?
- How will we measure it, and does that go beyond what Google’s report and our analytics already show?
- What’s the review date, and what happens to the line if the evidence doesn’t show up?
Take the Proposal Back With Better Questions
AI engines lean on what other people say about a business, and both Google and OpenAI describe the technical floor as ordinary website upkeep. The writing habits with measured gains already sit inside what a good editor produces. That leaves recognition as the work worth funding, though it grows slowly and favors larger names. No shortcut exists that Google’s guidance supports.
So diagnose the pitch before paying for the prescription. The managing partner can do it with a pen and the table above, and the four questions will surface more than the proposal volunteers. Compare the proposal with our team before deciding, and bring every line item along.
