Be the first sourceChatGPT cites.

Because the traffic number undercounts what is happening. When someone asks ChatGPT which app to use, the answer names two or three products and the person usually picks from that list without clicking anything. You only see the click if they go to your site afterward, and most of them go straight to the app store or type your name into Google, where it shows up as branded search or direct. So the referral line in GA4 is the tip of it. The bigger reason is timing. Engines settle on a default answer for a category and then keep repeating it, because the same sources keep getting cited. Getting into that default answer now, while the sources are still being chosen, costs a lot less than pushing someone else out of it in a year.

Show them a chain, not a score. Every month we report how many prompts in your set cited you, how many sessions came from AI referrers, how many of those sessions signed up or booked a demo, and what that is worth in your CRM. Those are numbers a finance person can check. If the chain has not moved after ninety days, that is a clear result and we will say so in the review rather than reframing it. We also show the work next to the numbers, so the board can see which page or source earned each gain instead of taking the trend on faith.

Voice search never changed where people made buying decisions, which is why it stayed a novelty. Answer engines already have. Google now puts an AI Overview above the organic results on a large share of queries, and Ahrefs measured a 58 percent drop in clicks to the top result when one is present. ChatGPT has hundreds of millions of weekly users and its search mode links out to sources on every answer. The thing that could go away is a particular product, but the layout of search, where a summary sits on top and cites a few sources, is now standard across Google, Bing, OpenAI, Perplexity and Anthropic. Planning for it is planning for how search works now.

Three ways, and you pick. We can run the whole program, classic search and AI search together. We can run only the AI search work and coordinate with your current agency through a shared backlog. Or we can do the audit, hand your team the plan, and step back. Our preference is to run both, because the same page rewrite that gets you cited in Perplexity also helps you rank in Google, and splitting that work across two vendors means two people editing the same pages and arguing about who moved the number. But if your agency relationship is good, we will slot in next to them and keep the handoffs clean.

Attend one forty-five minute recorded session every two weeks, answer questions in a shared Slack channel, and approve changes in your CMS before they go live. We write the drafts, stage them, chase the approvals and run the tooling. The time cost on your side is usually under two hours a week, most of it reading what we staged and saying yes or no. The one real ask up front is two hours with whoever knows the product best, so that the facts we put on the page are right the first time.

No, because what engines reward is not a writing style. They pull the clearest statement of a fact from the page that states it earliest and most plainly. That has nothing to do with tone. We move the direct answer up the page, add the specific facts an engine needs to repeat, such as prices, limits and dates, and put comparisons in a table where the question is a comparison. The words around those facts stay in your voice. If anything, the pages read more like a person talking and less like marketing copy, because padding gets cut.

This is usually the fastest win in the first month. A wrong claim about your pricing or features almost always traces back to one or two sources the model trusts, often a dated roundup, an old listing, or a forum thread. We find the source, get it corrected or outweighed, state the correct fact plainly on your own site, and then re-run the prompt in each engine until the answer changes. We keep a log of every claim, where it came from, and when it was fixed, so you can see the accuracy rate move month to month.

A tool tells you where you are not cited. It does not rewrite the page, fix the robots file, call the publication, or correct the listing. That part is labor, and it is where the result comes from. You can keep any tool you already pay for and our scorecard will line up with it, because we measure the same thing. What you are buying from us is the people who close the gap the tool shows you, with the measurement included so you do not need a second subscription to check our work.

About half of it is SEO with a new label. Google's AI Overviews lean heavily on pages that already rank, and ChatGPT's search results track Bing's top results closely, so ranking still matters. The other half is new. The unit of success is a citation inside an answer, not a position on a page. Third-party sources you do not own, like Reddit threads, review sites and trade press, carry more weight than they did in classic SEO, because engines cite them directly. And each engine retrieves from a different pool of sources, so a plan that works for one does not automatically work for the rest. We run both halves as one backlog so nobody has to referee between them.

Most of it overlaps. Moving the answer into the first sixty words of a page helps featured snippets and AI Overviews at the same time. Fixing crawl problems for OAI-SearchBot usually fixes them for Googlebot too. The parts that are different, like earning placements on the sources an engine cites and correcting wrong claims at their origin, do not touch your existing roadmap at all. Where there is a genuine conflict, say a page your SEO team wants to keep long and we want to restructure, we test it on one page and let the ledger decide.

ChatGPT's web search draws from Bing's index, and the pages it cites track Bing's top results closely. That means a brand that has ignored Bing for years can be invisible to ChatGPT even while it ranks well in Google. Setting up Bing Webmaster Tools, submitting through IndexNow, and checking how your key pages rank in Bing is week one work for any client who wants ChatGPT citations. It is cheap, almost nobody has done it, and it moves a result people care about.

We pull your server logs and look at what OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot, Bingbot and the others actually fetch, how often, and how fast your server responds to them. Then we compare that to what they cite. Most sites have at least one problem they did not know about: a robots rule or firewall blocking a bot, pages that need JavaScript to show their text, slow responses that cause the crawler to give up, or canonical tags pointing the wrong way. We fix those, add an llms.txt file, and keep watching the logs so a future deploy does not quietly break it again.

Yes. The standard list is GA4, Google Search Console, Bing Webmaster Tools, your CMS with staging rights so nothing publishes without approval, server logs or a regular log export, and a source field in your CRM so we can tie signups back to AI referrals. If your policy is view-only access for vendors, that works. We hand your team the exact changes as diffs and they apply them. It is slower but it is a reasonable way to start.

It matters most for the engines that read Google's index, which means AI Overviews, AI Mode and Gemini, and it helps every engine tell your brand apart from similarly named things. What it does not do is make a weak page citable. Schema is a label on the answer, not the answer. So we add it where it applies, such as product, organization, FAQ and how-to markup, after the page itself states the facts clearly. Doing it the other way around is how brands end up with perfect markup and no citations.

We take the prompt set we built with you, usually a few hundred questions your buyers ask, and run every one of them on every engine every day. For each run we record whether you were mentioned, whether you were cited with a link, which page was cited, and which competitors appeared. Share of answer is the percentage of runs where you appear, reported per engine and overall, with the trend. We also break it out by prompt tier, so you can see whether you are winning the high-intent questions or only the general ones.

Keep all of it. Our citation ledger sits next to your existing stack, and we read from your GA4, Search Console and CRM rather than replacing them. If one of your tools reports a different number than we do, we bring both to the next review and work out why, because that difference is usually a measurement setting rather than a real disagreement.

It depends on the category and on which prompts you are winning. Several 2026 datasets show visitors from AI referrers converting above organic for signups and purchases, because they arrive already knowing what they want. One more conservative dataset found B2B software companies converting slightly below organic. The honest answer is that the blended average is not useful. We tag every prompt by funnel stage, so you see conversion from comparison prompts separately from conversion from general questions, and we build the plan around the tier that pays.

Yes, as long as two things are in place. First, an AI referrer segment in GA4 that groups traffic from chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and the rest. Second, a source field in your CRM or product database that carries that segment through to the signup record. We also log which of your pages was cited for each prompt, so when a signup arrives from an AI referrer you can see which page earned the citation that produced it.

Some prompts you can earn organically and some you cannot, at least not quickly, because the engine only cites a fixed set of large sources for them. We flag those prompts in the backlog. As engines open up sponsored placements inside answers, that list becomes your buy list, and you already know the volume and the intent behind each prompt. In the meantime, the pages we build for citation tend to make good paid landing pages too, because they answer the question directly.

Yes, but through the web. Engines cite web pages, not app store listings, although they read store listings and use them as a source for facts about you. So we keep the App Store and Google Play listings accurate and current, and we make the web page that gets cited the shortest possible path to an install, with the store links high on the page and a clear statement of what the app does. We then track installs from AI referrers through your attribution tool the same way you track them from any other channel.

Accuracy fixes and rewrites of existing pages usually show up in the daily prompt runs within two to six weeks, because the engines re-fetch pages they already know about fairly often. New pages take longer to be discovered and trusted. Earned placements on third-party sources take longest, often two to three months, because you are waiting on someone else to publish. The audit sets a specific ninety-day target in your own units so you know what to expect by when.

Signups, installs or pipeline from AI-referred sessions, whichever is the unit your business runs on. Every other number on the scorecard exists to explain that one. Share of answer tells you whether you are in the answers. Citations by engine tells you where. Accuracy tells you whether the answer is helping or hurting. But the number that goes in the board deck is the one that ends in revenue.

No. The audit sorts your pages into three groups: pages that are cited and should be left alone, pages that are losing citations or getting none and should be rewritten, and prompts you have no page for at all. The first ninety days usually land on twenty to forty pages, a mix of rewrites and new pages, prioritized by how much prompt volume each one can win. The rest of your site stays as it is.

The answer stated plainly in the first sixty words, so an engine can lift it without reading the whole page. Specific facts with dates, numbers and names, because engines prefer claims they can repeat exactly. A comparison table when the question is a comparison. Headings that match how people phrase the question. And nothing standing between the crawler and the text, which means no login wall, no cookie overlay that hides content, and no reliance on JavaScript to render the words. Most pages fail on the first point and the last one.

Yes, and some clients prefer it that way. We hand your PR team the source map, which lists the publications, forums, review sites and listings each engine pulls from for your prompts, along with the angle most likely to get a placement on each. They run the outreach, we track whether the placement gets cited. If you do not have a PR team or they are busy, we run it ourselves.

For Perplexity and a meaningful share of ChatGPT answers, yes. Reddit threads are among the most cited sources for consumer questions, because engines treat them as unpaid opinions from real users. We do not create fake accounts or post reviews pretending to be customers. What we do is find the threads that are already being cited for your prompts, check whether the information about you in them is accurate and current, and make sure your company participates openly where that is welcome.

Yes. Engines extract facts, and tone is not a fact. We change the structure of the page, meaning where the answer sits and how specifically it is stated. Your team keeps the voice around it. In practice, pages tend to get shorter and more direct, which most brands consider an improvement once they see it.

Gemini and Google AI Overviews cite YouTube videos directly, often for how-to and comparison questions. For your highest-value prompts we make sure a short, accurate video answer exists, that its title and description match the question, and that a transcript lives on the related page so text-only engines get the same facts. You do not need a channel strategy for this. A handful of clear two-minute videos covers most of what the engines pull.

It is usually a better deal when you are small. In most categories the engines have not settled on a default answer yet, and the sources they cite are still thin. Being the first accurate, specific source for a question is cheap now and expensive later, once a bigger competitor has become the default and every engine keeps repeating them. We scope the prompt set to your size, so a small company is not paying to track prompts it cannot win.

Ninety days with no movement in share of answer, citations or accuracy for the prompts we agreed on. We have not had one, but we also do not have a long history yet, so we have built the contract around it. The first ninety days are fixed, and after that it is month to month, so if the ledger is flat you leave without an argument. You keep every page, every fix and the ledger itself.

A senior lead owns your engagement from the first call through every review, and that person is the one writing the plan and reading the ledger, not a junior account manager relaying messages. Behind them are specialists for crawler and technical fixes, content, and source outreach, pulled in as the backlog needs them. You will know who they are and they will be in your Slack channel.

Not in the same category at the same time. The reason is practical as much as ethical: the source map and prompt set we build for you is the core of the work, and it would be the same document for a direct competitor. We will tell you if a conflict comes up and let you decide.

Yes. A common first engagement is the citation audit plus a training block, where we walk your team through the ledger, the source map and the page patterns, and hand them the ninety-day plan to run themselves. Some teams run it from there, some come back for the retainer once they see how much of it is ongoing work. Either outcome is fine with us.

Those are the prompts we start with, because they decide the shortlist. Engines answer them by pulling from a small set of sources: roundup articles, Reddit threads, app store listings, and comparison pages on the apps' own sites. We make sure you are present in each of those with the facts the engine needs to recommend you, such as fees, which banks you work with, limits, and who the app is for. The specific prompts matter too. A general 'best app' question is hard to win. 'Best app to split rent with roommates when one person pays the landlord' is winnable, and the person asking it is closer to installing.

Yes, and it is usually the first thing we fix. The wrong claim almost always comes from one source: a roundup written before your last pricing change, an app store description nobody updated, or a thread where someone guessed. We correct the source where we can, state the right fact plainly on your own site in a place the engines re-fetch often, and re-run the prompt in each engine until the answer changes. Every claim goes in a log with the date it was fixed, so you can see accuracy improve.

Engines do not cite by brand size. They cite whoever answered the specific question best in a source they trust. A large brand with a general page loses to a focused app with a specific answer on narrow prompts, and narrow prompts are where most of the intent is. 'Split rent when one roommate moves out mid-lease' or 'pay rent with a credit card without a fee' are questions a big bank's page does not answer and yours can. We build the prompt set around those.

It works, but the website has to carry the answers, because engines cite web pages. Most app-first companies have a homepage, a pricing page and a blog, and none of them state the facts an engine needs. We typically build three to five pages written for retrieval: a how-it-works page, a fees page, a comparison page against the alternatives people name, and one or two pages for the specific situations your users ask about. That is usually enough to start getting cited.

Everything is staged and nothing goes live without your sign-off. We write to the claims your compliance team has already approved, and when we need a new claim, such as a specific fee or a bank partnership, we ask before we write it rather than after. If your review cycle is slow, we batch the changes so compliance reviews once every two weeks instead of every day.

Yes. Rent and roommate questions are often local, and engines answer them differently depending on the city in the prompt. The prompt set can include market-specific versions of each question, and we report share of answer by market so you can see where you are winning and where a local competitor or a local Reddit thread is beating you.

Both, run as one program. Google's AI Overviews still depend on ranking signals, and ChatGPT's search still mirrors Bing, so the SEO work has to happen. What changes is what we measure and optimize for. Classic SEO counts positions and clicks. We count citations inside answers and share of answer across engines, and we put more weight on third-party sources you do not own, because the engines cite those directly. One backlog covers both so nobody has to decide which discipline owns a page.

Not necessarily. We can run alongside them as the AI search layer, with a shared backlog so nobody edits the same page twice. Or we can run both classic search and AI search together. We usually prefer to run both, because the same page rewrites and the same earned sources move both results, and splitting the work creates handoffs that slow it down. If you like your current team, we will work next to them.

The citation audit is a fixed fee for two weeks, and you get the ledger, the source map and the ninety-day plan whether or not you continue. After that it is a monthly retainer scoped to the size of the prompt set, the number of engines we actively work, and the number of markets. We are flexible on scope. If the prompt set grows or shrinks, the fee adjusts with it, and we would rather right-size it than keep you on a plan you are not using.

Yes, and most clients do. It takes two weeks, it is a fixed fee, and you end up with the prompt set, the per-engine citation ledger, the source map and a ninety-day plan your team could run without us. If you continue, the retainer starts from that plan rather than from scratch.

Prompt tracking that runs your prompt set across every engine daily, crawler log analysis on your server logs, prompt volume estimates to size each question, and our own citation ledger that ties it together with your GA4 and CRM data. You never have to log into any of it. The monthly scorecard and the Slack channel are the interface. If you want direct access to the ledger, we will give it to you.

One forty-five minute recorded session every two weeks, a shared Slack channel for questions in between, and approvals in your CMS before anything publishes. Plan on under two hours a week. The recordings mean the people who could not attend can catch up, and decisions are written down in Slack so nobody has to remember what was agreed.

Accuracy fixes and rewrites of pages the engines already know usually show in the daily runs within two to six weeks. New pages take longer to be found and trusted. Earned placements on other people's sites take the longest, often two to three months. The audit sets a ninety-day target in your own units, so you know what should have moved by when.

Access to GA4, Google Search Console, Bing Webmaster Tools, your CMS with staging rights, server logs or a log export, and a source field in your CRM. On top of that, two hours with whoever knows the product best, so the facts we put on pages are right the first time. If any of those are hard to get, we start with what you have and add the rest as it comes.

ChatGPT, Perplexity, Google AI Overviews and AI Mode, Gemini, Claude, Microsoft Copilot, Grok and DeepSeek. The audit runs your prompt set on all of them so you can see where you stand everywhere. The retainer concentrates on the ones your buyers actually use, which for most consumer brands means ChatGPT, Google and Perplexity first, with the rest tracked and worked as volume justifies it.

Ninety days to start, then month to month. Ninety days is the minimum for the daily runs to show a trend you can trust, and for earned sources to begin landing. After that you can stop any month, and you keep every page, every fix and the ledger.