Most of what AI tells buyers about your product is written on pages you do not own. Here is what you can change.

/ 11 min read / By Faz

This autumn I asked Google AI Mode twelve questions that a buyer would ask about a named software product. Is it secure enough for our review. Who uses it and what did they get. How long will it take to roll out. Should we renew. Then I opened every source behind every answer and checked who wrote it.

There were 116 sources across the twelve answers. 37 of them were the product’s own pages. The other 79, about two in three, were written by someone else: competitors, consultancies, the company the buyer was leaving, people posting on LinkedIn, forum threads, review and benchmark sites.

That is the problem the last step of our method exists for. In the CITE Method it is E, earn citations in sources you do not own. This post is how that step actually works: which questions it matters for, who writes the pages, what you can realistically ask each of them for, and what a month of it looks like.

Your share depends on the question

The headline number hides the most useful finding. How much of the answer came from the vendor’s own site depended almost entirely on what the buyer asked.

Question type Tests Sources Vendor’s own pages Share
Security review 3 27 16 59%
Customer proof 3 27 12 44%
Rollout time 3 26 7 27%
Renewal 3 36 2 6%

When the buyer asked for facts only the vendor knows, where the data is stored, which certifications it holds, which customers use it, the vendor’s pages wrote most of the answer. The security tests and the proof tests are the detail.

When the buyer asked for a judgement, how long this will really take, whether to stay, how to pay less, the vendor’s pages nearly disappeared. In the three renewal answers, two of 36 sources were the vendors’ own, and both were pricing pages. The rollout tests and the renewal tests are the detail.

So the closer a question gets to a decision about money, the less your own website counts. That changes how you should spend effort. For questions in the top half of that table, the fix is mostly on your own site: publish the plain, specific page the engine is looking for. For questions in the bottom half, no page of yours will be enough on its own, because the engine is not looking for your opinion of yourself. That is where earning citations is the job.

Who writes the rest

Most advice about getting into AI answers is about one kind of page, the “best X tools” list. In these twelve answers, best-of lists were one kind of source among seven.

Rivals. Pricing pages, comparison pages, guides to migrating away from the product, a post about its hidden costs, a post titled around why it takes six months to implement. Rivals were the largest group in the renewal answers and in one of the rollout answers.

The provider being replaced. When the buyer said they were moving off one product, that company’s own guide to switching to the new one was cited, and the section on what slows the move down cited it first.

Partners, consultancies and agencies. A consultancy that runs migrations between two HR platforms. An agency’s post about a campaign it ran for a client using the product. An integration platform’s comparison of two tools. These people write guides because their clients ask them the same questions buyers ask the engine.

Other software companies in the category. AI support startups writing about a support platform’s limits, a sales tool writing a review of a bigger one, software companies explaining a rival’s pricing.

Practitioners and customers. A co-founder’s LinkedIn post about results, a practitioner’s post about the campaign they built, YouTube walkthroughs, LinkedIn posts about what a product’s plans include.

Communities. Reddit threads, and in one case the vendor’s own community forum, where a thread about how the product counts billable outcomes became the answer’s caveat.

Review sites, benchmarks and publications. A third-party benchmark of resolution rates, a technology publication, a McKinsey interview with a CEO.

Each of these needs a different approach, and for some of them the right approach is not outreach at all.

What you can ask each of them for

Start from the source map: the questions that matter, the pages each answer cites, and the exact sentence each page supplies. Then work down it by type.

Rivals: ask for factual corrections only, and expect little. You cannot pitch a competitor into recommending you. If a rival’s page states something about you that is factually wrong, an old price, a feature you now have, send a short, polite correction with a link to the evidence. Some will fix it. Most will not. The real response to a rival’s page is a page of your own on the same question, so the engine has a second source: an honest comparison, a page about the alternatives to you, a plain explanation of your pricing. Conceding where the rival is genuinely better is what makes that page credible enough to cite, which is the argument in winning comparison queries.

The provider being replaced: write the guide they will not. Their switching guide explains how hard leaving them is, because that is their job. Yours should be the most specific page about that exact move: what transfers, what has to be rebuilt, how long it takes, and what you do to help. You are not asking them for anything.

Partners, consultancies and agencies: give them material. This is the most underused group. They are already writing the guides the engine cites, and they want those guides to be accurate because their reputation rides on them. Do not ask them to mention you. Offer what makes their guide better: a current timeline, a migration checklist, an answer to a question their readers keep asking, a named person on your side who will check facts. If their guide is wrong about you, the correction is a favour to them. For integration partners, the asks are concrete: an accurate listing in their marketplace and a correct page in their documentation, which is the territory of integration pages.

Other software companies in the category: correct, and supply facts. They write about you when they compare themselves to you or explain your category. You will not change their conclusions. You can make sure the facts they use are current, and you can be the easiest company in the category to get an accurate answer from.

Customers and practitioners: help them publish their own results. The answers trusted named results more than vendor averages, and some of the most specific detail came from people posting in their own channels. Ask customers who have a real result whether they would share it themselves, on their own LinkedIn or blog, in their own words, with the numbers they are comfortable with. Help them get the numbers right. Do not write it for them, and do not pay for posts that do not say they were paid.

Communities: take part honestly, including on your own forum. On Reddit and similar communities, the only approach that survives is a named person, disclosed, being useful, sometimes recommending someone else. Why AI search cites Reddit covers why the shortcut backfires. Your own community forum counts too. If a thread there argues about something contentious in your pricing or limits, answer it plainly, because that thread may become the caveat in the answer.

Review sites, benchmarks and publications: get the category right, then earn the rest. On review sites the first thing to check is which category you are filed under, since that decides which questions you can appear in at all. The G2 post goes through it. Benchmark sites need real data, and a founder interview with a publication is slow to earn and worth having, but neither is something you can request by email next week.

A worked example

Here is one real source list, run through that playbook. The question asked how long a well-known conversation intelligence platform takes to roll out for a 40-person sales team, and what slows it down. The answer had ten sources.

Three were the vendor’s own: its homepage, its community forum, and a legal page on implementation services. The other seven came from five software companies selling to the same sales teams, several of them direct alternatives. The headline timeline, six to eight weeks to roll out and three to six months to real value, was attributed first to a rival’s post built around why the product takes six months and costs a lot to implement.

Taking each source in turn:

  • The rival’s “why it takes six months” post. Not a pitch target. Check it for factual errors and send a correction only if something is plainly wrong. The real response is on-site: a realistic timeline page of the vendor’s own, with phases, what the customer does in each, and what causes delays. Without it, the rival’s page is the only specific source on the question.
  • A second rival’s claim that one CRM integration takes far longer than another. If it is accurate, the vendor should say so first, with the reason and what it does to help. If it is not, a correction with evidence.
  • Delays drawn from reviews that rivals compiled. The underlying reviews sit on a review site. The move is to read what reviewers say about onboarding there, fix the causes that are real, and reply where a review is out of date.
  • The vendor’s own legal page, which supplied a 30-day deadline on using implementation help. That is a contract term doing customer communication. Explain it in plain words on a page buyers read.
  • The community forum thread about reps resisting the rollout. Already the vendor’s own ground. Answer it with the honest version and link the timeline page once it exists.

Of the five, two might justify a correction email, and only if something on the page is wrong. The rest is on-site work or answering on ground the vendor already owns. Earning citations is less about sending emails than about knowing, source by source, which of these moves applies.

How to write the ask

Most outreach fails because it asks for something the recipient has no reason to give. The asks that work are small, specific, and useful to the other person.

Quote the sentence. Say what is wrong or missing in one line. Link the evidence. Make one request. Do not ask for a link, and never offer payment for a change to editorial content. If there is no reply, send one follow-up a week later, and stop there.

A correction note can be four lines:

Hi Sam, your guide to moving off [product] says exports are only available on the Enterprise plan. That changed in March, it is now on every paid plan, here is the help page: [link]. Happy to answer anything else for the next update. Faz

That note will not get every page fixed. It is the version most likely to, because it costs the reader thirty seconds and makes their page more accurate.

What a month looks like

Unglamorous. A good month of this work is two corrections that land and one new source that describes you accurately. Nobody can promise a number, and anyone who promises a citation on a date is selling something they do not control.

Measure it the same way you found it. Re-run the same questions on a schedule, several times each, and watch two things. First the source list: a corrected page, a new partner guide, a customer’s post appearing among the sources. Then the answer itself. Source changes usually show up before the answer moves. Count a change in the answer only when it holds on at least two of three runs, because a single run is noise.

What not to do

No paid placements presented as editorial. No fake accounts, seeded threads, or customers posting reviews you wrote. No pressure on review sites. These are not only ethical lines. A page that gets caught doing any of them becomes a cited source about you, and it says exactly what you would not want, in someone else’s words. That is the same mechanism that writes a wrong label, set off by your own hand.

What I got wrong

When I rebuilt this method in September, I moved source-layer work to the front of everything, on the argument that the pages you do not own are what move answers. For the bottom half of the table above that is right. For the top half it is not. For security and proof questions, the vendors’ own pages wrote most of the answers, and the vendor that owned its security answer did it with plain help articles, not outreach. Sending outreach first on those questions would have been effort spent in the wrong place.

So the order is now: run the question, count whose pages wrote the answer, then decide whether the fix is on your site or off it. Earning citations is the lever for the questions where your own pages have stopped counting, and those happen to be the ones closest to the money.

These were twelve questions, run once each on one engine, and Google offered to rerun several of them without personalisation. The counts include every source the answer linked. Treat the shares as a pattern, not a measurement, and count your own.

Where this leaves you

For the questions that decide whether a buyer signs, stays or pays you less, most of the answer is written by other people. You cannot write it for them. You can find exactly which pages they are, correct what is wrong, give the people who write honest guides better material, help your customers say what happened in their own words, and publish your own plain page on every question where a rival currently has the only one.

If you want to see who writes the answers about you, and which of those pages can actually be changed, that is the first part of the engagement, and it runs on a documented method.

Want this run for your B2B SaaS?

Founding pricing for the first 5 clients. Methodology fully public. Month-to-month, cancel anytime.

Apply to work together

1 response

Leave a comment

Your email address will not be published. Required fields are marked *