This guide is for owners and marketers who want their business named when a customer asks an AI engine a question, instead of typing a search into Google and clicking through links. It covers two engines: Perplexity and Google AI Overviews. Those are the two engines CitedGEO measures today, and the only two this guide covers.
One limit before you start: AI answers are not fixed. The same question, asked twice, can name different companies. That is why every test below is run more than once.
Step 1: Measure your baseline before you change anything
You cannot tell if a fix worked if you do not know your starting point. A baseline is a small, fixed set of questions your buyers actually ask, run on each engine more than once, with the results written down.
A single run is a weak signal, because AI answers are not deterministic. The same prompt can name different brands on different runs, so a rate calculated over 3 runs is steadier than one screenshot. A worked example from CitedGEO's method: on a hypothetical set of 30 prompts run 3 times each on one engine (90 answers total), a brand named in 18 of them has a mention rate of 18 ÷ 90, or 20%. That is a hypothetical illustration of the formula, not a real result — your own numbers depend on your category and your prompts.
- ☐ Write 20 to 30 questions your buyers actually ask, in their own words, pulled from sales conversations, support tickets, search console queries phrased as questions, or community threads.
- ☐ Group the questions by intent: discovery ("best X for Y"), comparison ("A vs B"), problem ("how do I..."), and brand ("is A reliable?").
- ☐ Run each question 3 times on Perplexity and 3 times on Google AI Overviews, in a fresh session each time.
- ☐ For each answer, record: whether your brand is named, whether your domain is cited as a source, whether you are recommended as an option, and any wrong fact stated about you.
- ☐ Calculate your mention rate (answers naming you ÷ all answers) and citation rate (answers citing your domain ÷ all answers) per engine.
How to check it: Build a spreadsheet with one row per answer and these columns: date, engine, prompt, run number, brands named, domains cited, mention (yes/no), citation (yes/no), error (text). Fill it in by hand for your first 20 to 30 questions before you touch your site.
Step 2: Make sure the crawlers can actually reach your pages
An engine cannot cite a page it cannot read. Perplexity's crawler for search results is PerplexityBot; Perplexity states it is used to surface and link websites in search results and is not used to train its foundation models. If robots.txt, your CDN, or your firewall blocks it, your pages are invisible to that engine regardless of how well written they are.
Rendering is a separate problem. A crawler that does not run JavaScript sees only what is in the raw HTML. If your product details load in after the page renders, a crawler that does not execute scripts sees an empty page where your answer should be.
- ☐ Check robots.txt for a rule that blocks PerplexityBot on the pages you want cited. An explicit allow line looks like
User-agent: PerplexityBotfollowed byAllow: /. - ☐ Ask whoever manages your CDN or firewall to confirm bot-management rules are not silently blocking or challenging PerplexityBot, since robots.txt is only one gate.
- ☐ Fetch a key page with
curl -s https://yoursite.com/pageand search the output for a sentence that appears on the page. - ☐ If that sentence is missing from the curl output, treat the page as invisible to crawlers until the content is moved into the initial HTML.
How to check it: Run curl -s https://yoursite.com/your-key-page | grep "a distinctive sentence from that page". If the sentence does not print, the content is added by JavaScript after load and a non-rendering crawler will not see it.
Step 3: Find out which sources Perplexity already cites for your questions
Before you fix anything, find out where the engine currently gets its facts. In a September 2026 study of 76 Perplexity answers to "best SEO agency in [UK city]" questions, clutch.co appeared among the cited sources in 41 of the 76 answers — no other domain came close. The next four most-cited domains were themanifest.com (16 of 76), trustanalytica.com (8 of 76), agencies.semrush.com (7 of 76), and designrush.com (6 of 76). All five are directories or rating sites, not the agencies' own pages.
A separate study of 20 Perplexity answers to GEO and AEO agency questions found a different pattern: fewer directories, more of the agencies' own domains. The most-cited domain there appeared in 7 of the 20 answers, and clutch.co in 4. This suggests the mix of directories versus owned content varies by category, so you need to check your own.
- ☐ Run your baseline questions on Perplexity and write down every domain listed as a source in the answer, not just the brands named in the text.
- ☐ Sort the domains by how often each one appears across your question set.
- ☐ Note which of those domains are directories or review sites you do not control, and which are competitors' own pages.
- ☐ Flag any source domain where your business is missing or has an outdated listing.
How to check it: Add a "domains cited" column to your baseline spreadsheet from Step 1 and count how many of your 20 to 30 answers cite each domain. The domain with the highest count is where to start Step 4.
Step 4: Fix your listing on the sources that keep getting cited
Once you know which third-party pages Perplexity cites for your category, the next step is making sure those pages describe you correctly. If a directory or comparison page is missing your business, or lists an outdated price or address, the engine has no accurate version of you to cite.
Conflicting information across sources gives an engine a reason to be vague about you, or wrong.
- ☐ Claim your profile on the directory or review sites identified in Step 3, and complete every field.
- ☐ Correct outdated details: price, location, product names, founding facts.
- ☐ For comparison articles that omit you, contact the author with the specific facts they would need to include your business.
- ☐ For communities where your category is discussed, take part under your own name when the question is genuinely relevant, and disclose your affiliation.
- ☐ When an answer states something false about you, trace it to the cited source and request a correction there, not just on your own site.
How to check it: Re-run one of your baseline questions on Perplexity a few weeks after updating a listing, and check whether the source list for that answer now includes the corrected page.
Step 5: Make your own pages answer the question directly
When an engine writes an answer, it tends to quote or paraphrase a short passage rather than an entire page. If a page exists to answer "What does X cost?", the first sentence under that heading should state the price or pricing model, not build up to it.
Specific, dated statements are citable; vague claims are not. A stated example: "Supports exports to CSV and XLSX since version 4.2 (March 2026)" gives an engine something to quote. "Industry-leading export options" gives it nothing to verify or repeat.
- ☐ Rewrite key pages so the first sentence under each heading answers the question that heading asks.
- ☐ Put comparison tables in HTML text, not inside images, so an engine can read the cells.
- ☐ Write steps as numbered lists and definitions as single sentences.
- ☐ Add dates and sources to factual claims wherever the claim depends on data.
- ☐ Match structured data (Organization, Product or Service, FAQ, Article markup) to what the visible page actually says.
- ☐ Keep your company name, product names, category, and locations consistent across your site, structured data, and social profiles.
How to check it: Pick one page and read only the first sentence after each heading, out loud, with the rest of the paragraph covered. If that one sentence does not answer the question in the heading, rewrite it.
Step 6: Publish something Perplexity can point to
In the GEO/AEO study, each frequently named agency had something public that a source could cite: its own published ranking lists, a founder's weekly technical blog, a coined term for its method, a free guide, a book by its CEO, a keyword study on AI Overviews that is cited as a source in its own right, or a whitepaper. What these have in common is content specific enough, and public enough, for another page — or Perplexity itself — to point to.
This is not limited to marketing agencies. Any business can publish a dataset, a named method, or a detailed guide in its category, as long as it is specific enough to be worth citing rather than restating.
- ☐ Identify one piece of data, a process, or expertise your business already has that nobody else in your category has published.
- ☐ Turn it into a single, detailed, publicly accessible page: a study, a named framework, or a thorough how-to guide, not a short blog post that restates common knowledge.
- ☐ Give it a specific, memorable name so other sources can refer to it by that name.
- ☐ Do not use hidden instructions to AI models in white text or HTML comments; this is manipulation, it can be detected, and it risks the brand.
- ☐ Do not mass-produce near-identical pages for every city or keyword variant; they add nothing an engine needs to quote.
How to check it: Search Perplexity for a question your published piece answers and check whether your page, or a page describing it, appears in the cited sources.
Step 7: Re-measure on a schedule and watch for model updates
Run the same questions, with the same wording, on the same engines, on a set schedule, and compare the results with your baseline from Step 1. Keep a note of what you changed and when, so you can connect specific changes to specific results.
Answers move between runs even without any change on your part. In the GEO/AEO study, when the same ten questions were run twice each on Perplexity, the set of named agencies overlapped by 0.72 on average between the two runs, measured as Jaccard similarity, where 1.0 would mean an identical list both times. That is why a single screenshot of one answer is a data point, not a trend, and why each re-measurement also needs several runs per question.
AI vendors also update their models and search systems, sometimes without public notice, and this can move every number in a report overnight for reasons unrelated to your work.
- ☐ Set a fixed date to re-run your full baseline question set, using the exact same wording as the first run.
- ☐ Run each question at least 3 times per engine, as in Step 1, not once.
- ☐ Compare mention rate and citation rate against the baseline, per engine, rather than blending both engines into one score.
- ☐ If you notice a sudden shift across most of your questions at once, check the vendor's release notes for a model update before assuming your changes caused it.
- ☐ Mark a line on your tracking chart at any point where a model update happened, and do not compare numbers across that line as if nothing changed.
- ☐ Re-check which domains get cited for your questions after a model update, since the source mix can change.
How to check it: Compare two dated rows in your spreadsheet from Step 1 for the same question, three months apart. If the mention rate changed and no model update happened in between, the change is more likely tied to your own work.
Checklist summary
- ☐ Write 20 to 30 buyer questions, grouped by intent (discovery, comparison, problem, brand).
- ☐ Run each question 3 times on Perplexity and 3 times on Google AI Overviews and record the results in a spreadsheet.
- ☐ Calculate mention rate and citation rate per engine from that baseline.
- ☐ Check robots.txt for rules blocking PerplexityBot on key pages.
- ☐ Confirm your CDN or firewall is not silently blocking or challenging PerplexityBot.
- ☐ Fetch a key page with curl and confirm its main text appears in the raw HTML, not only after JavaScript runs.
- ☐ List every domain Perplexity cites as a source across your baseline answers, sorted by frequency.
- ☐ Flag source domains where your business is missing or has outdated information.
- ☐ Claim and complete your profile on the directories and review sites that keep getting cited.
- ☐ Contact authors of comparison articles that omit you, with the facts needed to include you.
- ☐ Trace any false statement about you to its cited source and request a correction there.
- ☐ Rewrite key pages so the first sentence after each heading directly answers that heading's question.
- ☐ Put comparison tables in HTML, use numbered lists for steps, and add dates and sources to factual claims.
- ☐ Match your structured data to what the visible page actually says.
- ☐ Publish one detailed, specifically named, citable piece of content: a dataset, a method, or a thorough guide.
- ☐ Avoid hidden text instructions to AI models and mass-produced near-duplicate pages.
- ☐ Set a fixed re-measurement date and repeat the same questions, 3 runs per engine, on that date.
- ☐ Mark any model update as a break in your trend line and do not compare numbers across it.
Method notes and limits
The numbers in this guide come from CitedGEO's own published studies and method pages, not from a client project. The clutch.co and agency-naming figures come from two small Perplexity-only studies run in September 2026: 76 answers for the UK local SEO set, 20 answers for the GEO/AEO set. Sources counted are the links visible in the answer text, not each answer's full source list, so the counts understate how often a domain was consulted but left uncited.
No one can guarantee that an AI engine will name your business. The vendor decides what its answer says, and the same question can produce a different answer an hour later. What you can control is whether your pages can be read, whether they answer the question directly, and whether the sources these engines already cite describe you correctly.
About CitedGEO
CitedGEO is a Generative Engine Optimization agency that measures how often brands are named and cited in AI answers, and publishes its measurement method openly. Its free GEO audit runs 10 buyer questions through Perplexity and Google AI Overviews, 3 times each, producing 60 recorded answers, and shows where your business appears, where competitors appear, and what to fix first. Details are at /free-audit/.