Most earned-media plans aimed at AI visibility start with a target list somebody assembled from reputation: the trade titles the team already reads, the review platform the sales team complains about, two big publications nobody has a route into. Then the plan runs for a quarter and nobody can say whether any of it touched what assistants say about the company.
The order is wrong. The list should come out of measurement — the sources that actually appear in answers to the prompts your buyers use — and every route into those sources should be checked against that platform's own published rules before anyone pitches anything. This post covers that sequence, and it spends a full section on where the line sits, because this is the part of AI visibility work where the shortcuts are cheapest and the consequences are worst.
Start from measurement, not from a media list
You need data before you need a plan. The input is a prompt set that reflects how buyers in your category actually ask, run repeatedly across the assistants your buyers use, with the cited sources recorded for every answer.
The mechanics of collecting that — sampling, repeat runs, coding the results — are covered in full in which sources decide your category. The short version:
-
Build a prompt set from real buyer language, not from keyword tools.
-
Run each prompt multiple times per model. Answers vary between runs, and a source that appears once in ten runs is not a finding.
-
Record the domain, the URL, the page type, and the date for every citation.
-
Do this before you spend anything, so you have a baseline to compare against later.
What you end up with is a frequency table of domains, not a media list. It will contain surprises. It usually contains fewer marquee publications than the team expected and more niche, structured, comparison-shaped pages than anyone wanted to hear about.
Two warnings about the data itself. Citation behaviour differs substantially between assistants, and a source that dominates one may be absent from another. And phrasing moves the mix: "best X for small teams" and "X alternatives" can pull from different corners of the web. Do not collapse it all into one ranked list before you have looked at the splits.
Categorizing the sources you find
Raw domains are not actionable. Group them by what kind of thing they are, because the route in is determined by the type, not by the name.
| Source type | Typical examples in a frequency table | Route in | Effort |
|---|---|---|---|
| Editorial / trade press | Category publications, journalist-written roundups | Pitch, expert commentary, data | High, slow, no control over outcome |
| Review and comparison platforms | Software directories, category review sites | Claim and complete your listing; ask customers to review under the platform's rules | Medium, ongoing |
| Community discussion | Forums, Q&A sites, subject-specific communities | Participate under your real identity where the platform permits it | Low cost, high care required |
| Product listings and marketplaces | App stores, integration directories, partner catalogues | Complete and maintain the listing | Low, mostly one-off |
| Reference and documentation | Encyclopaedic references, standards bodies, official docs | Usually not a route in at all — accuracy corrections only | Low, narrow |
| Your own properties | Your site, docs, support content | Publishing and structure work | Fully within your control |
| Aggregators and scrapers | Content-thin sites republishing others | No legitimate route; ignore | — |
The last two rows matter as much as the first. If your own domain is already cited frequently, the highest-return work may be improving what those pages say rather than chasing third parties. And if a scraper aggregate is showing up, the fix is upstream: correct the original source it copied.
Legitimate routes: editorial, contributed, community participation, product listings
Before any route: read the platform's own rules first. Every category below has platforms with published guidelines about who may participate, under what identity, and with what disclosure. Those rules are the binding constraint, they differ platform to platform, and they change. Nothing in this section overrides them, and none of it is a substitute for reading the current version of the specific policy.
Editorial coverage
The rules that apply. Publications have editorial standards, and most reputable ones publish them: disclosure of commercial relationships, separation of sponsored content from editorial, and a stated policy on whether contributed pieces are accepted and how they are labelled. Read the masthead and the contributor guidelines before pitching.
What you can do. Pitch a story with something in it the journalist cannot get elsewhere — original data, a customer situation you have permission to describe, informed commentary on something happening in the category. Offer a named expert who can actually speak. Respond fast when a reporter asks.
What you cannot do. You cannot buy an editorial mention and have it remain editorial. Paid placement that reads as editorial without a label breaches most publications' policies and, in advertising terms, is a disclosure problem in many jurisdictions. If money changes hands, the label goes on.
Honest expectation. You control the pitch. You control nothing else. A pitch can be perfect and still be declined for reasons that have nothing to do with you.
Contributed and byline articles
The rules that apply. Where a publication accepts contributed pieces, it will usually specify authorship (a real named person, not the brand), promotional limits, and required disclosure of the author's employer. Some publications do not accept contributions at all.
What you can do. Write under a real byline, disclose the affiliation clearly, and make the piece useful independently of your product.
What you cannot do. Ghost-written pieces attributed to someone who did not write them, or pieces that hide the commercial affiliation, breach standard contributor terms.
Community participation
This is where the highest volume of bad advice in this field lives.
The rules that apply. Community platforms have explicit self-promotion policies, and they vary sharply: some prohibit any commercial participation, some allow it with disclosure, some allow it only in designated threads or from verified accounts. Many prohibit multiple accounts outright. Read the specific community's rules, not a general summary of "how communities work."What you can do. Participate as yourself, with your affiliation visible, answering questions in your area of competence, including when the honest answer is that a competitor fits the person better.
What you cannot do. Anything that hides who is speaking. This is covered in detail below.
Review platforms and product listings
The rules that apply. Review platforms publish rules on soliciting reviews, and they are strict and specific: typically no incentives conditional on sentiment, no filtering of who is asked based on expected rating, no employee or agency reviews posed as customers, and required disclosure of any incentive that is permitted. Advertising regulators in several jurisdictions also have rules covering incentivized and fake reviews. Read both the platform policy and the applicable regulator guidance before you run any review campaign.
What you can do. Claim your listing, keep it accurate, and ask customers to review — all customers, not a screened subset — through whatever mechanism the platform sanctions.
What you cannot do. Reward positive reviews, suppress negative ones, or generate reviews from anyone who is not a customer. Reviews and third-party coverage goes deeper on what review platforms plausibly contribute and where the bright lines sit.
The line: what is participation and what is manipulation
Scope note. This article is general educational and measurement guidance, not legal, tax, financial, security, or platform-specific advice. Current law, first-party platform rules, the facts of the real case, and qualified professional advice control any consequential decision. The line is identity and disclosure. If a reasonable person reading the placement could tell who is speaking and what their interest is, it is participation. If the value of the tactic depends on that being concealed, it is manipulation. Everything else follows from that test.
Concrete examples on both sides:
| Participation | Manipulation |
|---|---|
| Answering a community question with your affiliation visible in your profile and in the reply | Posting the same answer from accounts that do not disclose the affiliation |
| Asking every recent customer for a review through the platform's sanctioned flow | Asking only customers you expect to be positive, or offering anything conditional on the rating |
| Pitching a journalist with original data and letting them write what they conclude | Paying for a mention that runs without a paid label |
| Writing a contributed piece under your own name with employer disclosed | Ghost-writing under an unaffiliated person's byline |
| Correcting a factual error about your company on a reference page through its documented process | Editing a reference page about yourself while presenting as a neutral party |
| Publishing a comparison on your own site that states its bias openly | Publishing "independent" comparison sites you own without disclosing ownership |
| Maintaining an accurate directory listing | Creating listings for products or locations that do not exist |
Note what the manipulation column has in common. None of those items is bad because it is aggressive. Each is bad because it works only if the audience is wrong about who is talking.
Three reasons this matters commercially, beyond the ethics:
-
Platforms detect it and act. Review platforms and communities run detection on coordinated posting, and enforcement removes the placement along with the account history behind it.
-
Regulators cover it. Fake and undisclosed-incentive reviews sit inside advertising and consumer-protection law in multiple jurisdictions. This post is general information, not legal advice; get qualified counsel before running anything you are unsure about.
-
It contaminates your own measurement. Placements you manufactured tell you nothing about whether the category actually responds, so you lose the ability to learn from your own programme.
We cover the mechanics of astroturfing and review fraud defensively only — how they are detected and what they cost — in review fraud and astroturfing. You will not find a method here, and you should be wary of anyone who offers one.
Prioritizing by observed citation frequency and effort
Now combine the two things you have: how often a source appeared in your own runs, and how hard the legitimate route into it is.
A workable ordering:
- Frequently cited, and you control it. Your own pages appearing in answers. Fix the content first; it is the only category where you decide the outcome.
- Frequently cited, low-effort legitimate route. Unclaimed directory listings, incomplete profiles, communities where you already have credible people. Do these next because the cost is hours, not quarters.
- Frequently cited, high-effort route. Trade press and major review platforms. Worth a sustained programme, with the understanding that outcomes are not within your control.
- Occasionally cited, low effort. Batch these. Do not build a programme around them.
- Occasionally cited, high effort. Decline explicitly and write down why, so it does not come back next quarter.
- No legitimate route. Scraper aggregates, platforms whose rules prohibit your participation. Leave them alone.
Two constraints on the ordering. Concentration matters: if three domains account for most citations in your category, effort spent outside them is speculative. And volatility matters: a source that dominated one month's runs and vanished the next is not a stable target, and you should re-measure before committing budget.
Measuring whether a placement changed anything
A placement is an output, not an outcome. Treat every one as a change you now have to test.
-
Baseline before. You need the pre-placement citation and mention data for the same prompt set. Without it there is nothing to compare against.
-
Wait. Systems index and refresh on their own schedules, and there is no published, reliable figure for how long a new page takes to appear in assistant citations. Anyone quoting you a number should be asked where it came from.
-
Re-run the same prompt set, same models, same number of runs. Changing the instrument between measurements makes the comparison meaningless.
-
Expect noise. Answers vary run to run. A shift smaller than the variance you already measured is not evidence of anything.
-
Do not claim causation from one placement. Other things happened in the same window: your own publishing, competitors' activity, model updates. The correct claim is "we placed X, and citation frequency for source type Y moved from A to B across N runs on these dates," with the confounds named.
An honest programme accumulates weak evidence across many placements and slowly learns which source types respond in your category. That is a real result. A single placement followed by a good month is not, and presenting it as one will cost you credibility the first time the pattern fails to repeat.
In Plain English
Do not build an outreach list from reputation. Measure which sources appear in answers to your buyers' actual questions, group them by type, and pick routes based on that data.
Before pitching, participating, or asking for reviews anywhere, read that platform's own participation rules — they differ, they are binding, and they change. The line between participation and manipulation is identity and disclosure: if the tactic only works when people cannot tell who is speaking, it is over the line, and both platforms and regulators treat it that way. This is general information, not legal advice.
After any placement, re-measure with the same prompt set and the same models, and describe what moved without claiming you caused it. This helps most if you have earned-media budget and no evidence base for spending it. It helps least if you have not measured anything yet — start there instead, via what actually changes recommendations.
Use the research room protocol to evaluate source types and document how you measure them.
Primary-source verification
Checked July 28, 2026. The operational and factual guidance above was reviewed against the primary references below. Provider features, prices, policies, interfaces, jurisdictions, and account-specific results can change; recheck the controlling first-party document and your own account or measured data before acting.
- Google Search: AI features and your website - How Google describes eligibility, preview controls, and appearance in its AI search features.
- Google: common crawlers and special-case user agents - Current Googlebot and Google-Extended identities and controls.
- OpenAI: overview of OpenAI crawlers - Current GPTBot, OAI-SearchBot, and ChatGPT-User descriptions and publisher controls.
- Anthropic: web-crawler controls - Current ClaudeBot, Claude-SearchBot, and Claude-User descriptions and controls.
- Perplexity: crawler and user-agent documentation - Current PerplexityBot and Perplexity-User descriptions and controls.
- FTC: Endorsement Guides Q&A - Current U.S. regulator guidance on endorsements, disclosures, and reviews.
- FTC: Consumer Reviews and Testimonials Rule Q&A - Current U.S. rule guidance on deceptive review practices.