Generative Engine Optimization (GEO): How To Get Cited By Perplexity AI

The age of the ten blue links is drawing to a close.

With the continual change in user intent away from traditional search engines and towards generative answer engines, Organic Traffic Strategies are currently fragmenting.

The modern user has no desire to search through ad-laden recipe websites or cumbersome corporate how-to pages to locate a single actionable piece of information; they want it now.

Perplexity is positioning itself as the driving force behind the significant evolution occurring in the information retrieval process.

For Brands, Publishers, Marketers, etc., this imposes an entirely new standard for obtaining visibility.

Simply relying upon the traditional methodologies of keyword density and backlink velocity to achieve organic visibility for your content is no longer viable.

You must be strategically optimizing your content for the AI powered model crawling the internet in real time. This is what is referred to as Generative Engine Optimization (GEO).

Unless your Pages are explicitly structured for ingestion, summarization and citation by AI models, your traffic through referrals will eventually experience decline.

The citation playbook for Perplexity

To effectively optimize for generative search, you will need to disassemble and rebuild all of the old school SEO Tactics.

The Citation Playbook for AI search visibility places a strong emphasis on structure, clarity and unobstructed accessibility.

Reviewing our top-performing AI-cited contents reveals several mandatory requirements:

• Bot Access is Required: If your robots.txt file is preventing bots from crawling through your pages, you are effectively invisible to Perplexity.
• Freshness Overrides Everything Else: Perplexity displays a significant bias towards recent updates. Pages that are updated within the last week regularly outperform highly authoritative pages that have not changed in three years.
• Structure Determines Extraction: Models are able to extract Answers most efficiently from Pages that are structured with Question Based Headings immediately followed by Dense, Direct Answers.

The amount of specific, named, quantifiable, and statistically analyzed information contained within a sentence determine if it gets cited (as opposed to vague marketing content).

To understand citation generation, you need to know how Perplexity retrieves resources.

Perplexity isn't just a Large Language Model that spits out memorized data; Perplexity is a Real-Time Retrieval Augmenting Generative system.

When a user makes a request, Perplexity first understands what the user wants to do with it, and then performs a live search of the web to retrieve all relevant content, after which it reads all of the unformatted content that it finds, summarizes it into a single answer, and provides citations based on which resource provided the most accurate and reliable information.

The foundation of any visibility on Perplexity starts with Technical Accessibility, which Perplexity creates using its own bot, PerplexityBot.

In late 2023, many publishers panicked by blocking all AI bots so that their data would not be used in training Large Language Models.

However, if you do this for PerplexityBot, you effectively close the door to visibility.

If PerplexityBot is blocked, Perplexity cannot access your page in real time to produce an answer to a user's request. Therefore, by blocking PerplexityBot, you are forfeiting all potential referral traffic from Perplexity.

Therefore, the first step in any GEO campaign is to allow this specific user-agent.

While "old" search engines still rely heavily on backlinks as a measure of Trust, "new" generative search engines rely primarily on recentness as the primary measure of Trust.

For example, when a user asks about the "best CRM software," the "old" search engines would display all of the oldest pages that have the highest amount of backlinks, whereas "new" generative search engines will look for the most recent information available about CRM software, as opposed to what is the most backlinked.

The most recently revised guide, published three days ago, has a more accurate and scientifically-based set of premises upon which an LLM will be constructed, than does the extremely comprehensive, authoritatively written guide from 2021.

The routine act of removing stale content and updating data and information, is among the leading forces behind LLM citation.

Above all else, LLMs function as predictive engines, and therefore, they use the principle of ‘high density’ information.

Summarizing information is performed by LLMs by taking advantage of data-rich sections of text.

In addition to reducing the number of distractions for an LLM, excessive ‘fluff’ also inhibits the LLM’s ability to pull information from a page.

Long, winding history lessons (about any subject area) take away from the purposeful transfer of information—the less semantic value there is, the less relevant and logically predictable the continuation of text.

LLM engines have an affinity toward named entities, specific numerical values, dated dates, and Technical definitions.

The pages an LLM produces regarding a subject area, must contain all of these betting elements.

Pages formatted for LLM optimization, essentially, eliminate the “fluff” and prioritize factual references.

Ideas on how to create high-density pages for LLMs

An understanding of what an LLM is looking for, is only one half of the battle.

The manner in which you format your web page will support your overall opportunity for easy-extraction of information via an LLM.

Constructing web pages to resemble ‘clean’ data feeds.

Directly answer questions with the H2s on the page

The best way to format your web page is to replicate the process by which an AI searches.

Users will often have a question for the AI, when they input that question into the AI, the AI will have a predetermined format to find an answer for the user’s question.

Formulate your H2 headings as question-styled headings. Do not be clever or creative with your headings.

Place a TL;DR (too-long, didn’t-read) above the H2.

The TL;DR should answer directly the question in the heading, without getting distracted with unnecessary introductions or related comments. This is your prime extraction opportunity for the Generative Engine.

When you have the direct answer to your question, write a few more paragraphs as needed to explain your solution in more detail.

Citing sources at the sentence level

Citation occurs at the individual sentence level.

The search app doesn't cite all the information on a page, it only cites the specific fact that it obtained from the specific individual sentence.

As a result, in order to have a higher citation rate, you should write in a direct and objective style.

Eliminate any conditional phrases and all marketing descriptors from your text.

Each paragraph you write should contain one or more verifiable facts, statistics or named references.

Examples of traditional vs. AI optimized citable paragraphs

The following examples illustrate the difference between traditional marketing copy writing and writing for an AI optimized model.

The following example does not contain any verifiable information:

"We offer very good digital marketing services to many different types of companies. Our staff makes every effort to increase your website's search engine position and to increase the organic traffic that your website receives."

The model would not extract anything from this example; this is because there are absolutely no verifiable facts contained within the paragraph.

The following paragraph contains the same basic information but is optimized for Perplexity citations:

"Acme Agency offers locally-focused SEO solutions to B2B SaaS Providers. In 2023, average organic traffic increases of 42% over six months were experienced by clients whose campaigns utilized 'Entity Based Schema Markup' and who also had Technical Site Audits conducted."

The second example is a strong candidate for citation.

It clearly states which services we're providing, who we provide them for, what techniques we're using, and finally, provides a specific statistical outcome.

Annotated page template for page level content

If you're going to create new page content from scratch, the general flow of the page's content should follow this model.

Use an H1 header that describes the page "theses". Just underneath the H1 header, write a short executive summary block of the main takeaways for the entire page.

Organizing your body copy with headings (H2) is an essential step in attaining citations via Perplexity. As will be detailed in the corresponding bullet point sections below, AI models parse itemized lists very well.

Technical checklists for getting cited by Perplexity

To optimise your strategy, execution is paramount.

Ensure that your technical website foundation conveys trustworthiness, structure and authority as signals to parsing algorithms.

Unblocking crawlers and indexing

To unblock crawlers and index your site, go to your robots.txt file and ensure that User-agent: PerplexityBot can access your website without being blocked.

In addition to enabling the bot to access your site, you should also confirm that your site has a flat architecture and is easy to navigate.

If your site relies heavily on client-side JavaScript to deliver core content, you may experience delayed or absent indexing by artificial intelligence tools.

The best way to ensure immediate ingestion of fact-dense, critical text, is to serve this information as plain HTML, whenever you can, during the actual crawl of the bot.

Schema markup: Generating context and building connections

Schema markup acts as a bridge between structured and unstructured data, allowing generative engines to place content within context based on the structure of the schema markup.

The generative engine has access to structured data, which is created through the use of schema markup, and can maintain and establish connections between differing attributes of schema markup to generate more quality citation links.

Instead of using only the most basic schema markup (i.e., "website" schema), you should create a comprehensive schema markup that is highly detailed and targeted.

For example, create and deploy the FAQPage schema to explicitly define your questions and answers — allowing both search engines and users to locate and understand where to find answers to their questions.

Use either the Article or NewsArticle schema in order to convey the timestamps of your published content (i.e., creation date and last updated date), which directly corresponds to the recency factor considered by generative engines.

Consider using the Organization and Person schema to establish both the authority of your company and the expertise of your authors and contributors.

Authority & relevance: Building topical clusters & topical authority

When building topical clusters, it is important to note that, generally speaking, AI engines do not generally cite individual blog posts that are not also part of a larger topical cluster of related topics.

Thus, the majority of your citations will come from well-optimised topical clusters with extensive, topical authority.

Establishing topical clusters - In order to reach "Cloud Server Migration" authority status, you will need to create one central pillar page tied to dozens of highly technical subpages that describe specific elements of this process in detail.

The interconnected web of internal links signals to the algorithm that your domain has extensive, focus-based expertise associated with that particular topic.

Workflow Process - The bridge between theory and execution is generally where teams fail. Therefore, you must implement a disciplined method for auditing, refreshing, and monitoring your content for visibility on AI-based search engines.

Phase one - Initial assessment of content

First, identify your top value pages (generally referred to as core-service or homepage). Examples include services offered, comparison guides, and flagship reports.

Second, review your highest-value pages according to GEO's guidelines. Have the headings been formulated as questions? Does the first paragraph under each heading provide a direct, factual, and succinct answer? Is the content no longer recruitable?

Any page that hasn't received an update (refresh) within the past six months should be flagged to be updated; and continue by removing the merchant's promotional language, adding new statistical data, using current methodology, and providing accurate schema markup.

Phase two - Establish a monitoring system

At this time, SEO will be tracked in various places; however, Google Analytics (GA) metrics cannot measure Perplexity impressions.

To measure citations that derive from this new model of search, one must depend on server log analyses and referral traffic. Look for unusually large amounts of referral traffic coming from perplexity.ai, documenting which landing pages received that traffic.

In addition, utilize personnel with specialized experience in monitoring citation visibility with new GEO-tracking technology that simulates search queries over the course of time.

Perform these checks on a weekly basis to align content refreshes with the speed at which citations are generated.

AI search visibility's most common friction points

Shifting towards an AI-centric optimization approach will bring substantial friction into day-to-day operations.

If you can detect these points of congestion early, you can avoid wasting money and time.

Updating frequency vs. resource allocation

The extreme focus on "how fresh" content is creates a huge operational burden on resources.

An editorial team cannot keep up the pace of rewriting the entire library of content every quarter.

The answer is aggressive triage. Do not make the attempt to refresh every piece of content posted on your site.

Instead, create a list of the 10-20 most commercially successful pages or pages that are recognized as authoritative for your brand. Create an ongoing schedule to refresh those specific pages on a regular basis.

Add in new data points, change the last modified schema dates, and enhance the TLDR text. Allow less commercially successful pages to age naturally.

Technical challenges for non-developer teams

There is a technical challenge to implementing advanced implementations of structured schema markup and reviewing server logs for PerplexityBot crawl activity.

For smaller companies with limited resources or local agencies without dedicated software developers, this becomes a significant point of congestion.

To overcome this friction point, utilize no-code schema generators and plugins that add structured data automatically to your web pages.

If reviewing server logs is not feasible, focus on just the on-page elements that are structurally sound: proper H2 hierarchy, dense paragraph structure, and frequent content updating. The structure and format alone will weigh heavily on the citation process.

Examples of over-optimization

In the same way as in traditional SEO, it’s also easy to manipulate the process by stuffing artificial statistics onto your pages, making up expert quotes and/or creating masses of low-quality AI-written pages in order to create large clusters of fake topics.

Generative models have become increasingly skilled at detecting semantic spam as a result of this.

Content that appears to be written as a robot listing unrelated facts are passed over in favour of generative models that detect and retrieve content that provides readers with a high density of information that also reads with an editorial flow familiar to human readers.

You should create content for the purpose of being extracted by generative models, but you should also create factual-based content that provides your audience with factual information based upon reality.

Understanding Perplexity optimisation

The way in which search works is dramatically changing, and we are transitioning from a Page Retrieval-Based approach to a Fact Retrieval Based approach.

To achieve visibility in this new environment, you will need to be extremely clear.

This means that you will need to eliminate the wordy, and filled with fluff, marketing copy of the last decade. This marketing copy should be replaced with structured, dense, and very consumable data for the user.

Allow search bots to crawl your site. Consistently update your fundamental, core pages. Provide answers to the questions that the user has as quickly as possible. Backup and validate the information you provide by providing verifiable sources.

Those that adapt their content architecture to the needs of generative models will benefit from increased numbers of high-intent referrers.

Those that continue to use keyword-stuffed, long-form articles will become increasingly irrelevant in the AI-generated consensus as time goes on.

Frequently asked questions

How long until the citations show up in Perplexity?

Citations for a given page can show up almost instantly once PerplexityBot has indexed that page.

Based on our testing and tracking, we have typically seen citation visibility often move/shift within the 48 to 72 hour range after a page has been updated and the crawl verified - assuming that the targeted content perfectly matches User Intent.

Does Perplexity have a preference for certain types of domains?

Yes, Perplexity demonstrates a strong preference towards domains that possess existing topical authority and high trust signals.

When compared to generic, low-quality affiliate sites, academic publishers, established industry blogs with clear author bylines, and highly structured technical websites tend to perform significantly better in terms of Perplexity citations.

Trust is one of the major contributors to the reduction of AI Hallucinations in RAG systems.

Are Generative Engine Optimization (GEO) efforts fundamentally different than traditional SEO?

Yes, they are fundamentally different.

In traditional SEO, off-page signals (such as backlink profile/s) and keyword density (exact match) carry the most weight. GEO has a greater emphasis on the on-page architecture, sentence-level factual density, content freshness, and schema markup.

Pages that have a low ranking in Google may be the primary source of citations in Perplexity, as long as they have an architecture that is optimised for AI Extraction.

What will happen if my 'robots.txt' file blocks AI Crawlers?

If PerplexityBot is blocked in your robots.txt file, your live pages cannot be accessed via real time User Queries to Perplexity.

Although the underlying LLM may have accumulated some historical knowledge of your brand from its original training dataset, it would not be able to cite your current content, link to your domain or extract any new factual information related to your business.

By blocking Perplexity, you completely remove yourself from the referring traffic ecosystem of the platform.