Boost your skills with Growth Memo’s weekly expert insights. Subscribe for free!
Here’s what I’m covering this week: How to build user personas for SEO from data you already have on hand.
You can’t treat personas as a “brand exercise” anymore.
In the AI-search era, prompts don’t just tell you what users want; they reveal who’s asking and under what constraints.
If your pages don’t match the person behind the query and connect with them quickly – their role, risks, and concerns they have, and the proof they require to resolve the intent – you’re likely not going to win the click or the conversion.
It’s time to not only pay attention and listen to your customers, but also optimize for their behavioral patterns.
Search used to be simple: queries = intent. You matched a keyword to a page and called it a day.
Personas were a nice-to-have, often useful for ads and creative or UX decisions, but mostly considered irrelevant by most to organic visibility or growth.
Not anymore.
Longer prompts and personalized results don’t just express what someone wants; they also expose who they are and the constraints they’re operating under.
AIOs and AI chats act as a preview layer and borrow trust from known brands. However, blue links still close when your content speaks to the person behind the prompt.
If that sounds like hard work, it is. And it’s why most teams stall implementing search personas across their strategy.
- Personas can feel expensive, generic, academic, or agency-driven.
- The old persona PDFs your brand invested in 3-5 years ago are dated – or missing entirely.
- The resources, time, and knowledge it takes to build user personas are still significant blockers to getting the work done.
In this memo, I’ll show you how to build lean, practical, LLM-ready user personas for SEO – using the data you already have, shaped by real behavioral insights – so your pages are chosen when it counts.
While there are a few ways you could do this, and several really excellent articles out there on SEO personas this past year, this is the approach I take with my clients.
Most legacy persona decks were built for branding, not for search operators.
They don’t tell your writers, SEOs, or PMs what to do next, so they get ignored by your team after they’re created.
Mistake #1: Demographics ≠ Decisions
Classic user personas for SEO and marketing overfocused on demographics, which can give some surface-level insights into stereotypical behavior for certain groups.
But demographics don’t necessarily help your brand stand out against your competitors. And demographics don’t offer you the full picture.
Mistake #2: A Static PDF Or Shared Doc Ages Fast
If your personas were created once and never reanalyzed or updated again, it’s likely they got lost in G: Drive or Dropbox purgatory.
If there’s no owner working to ensure they’re implemented across production, there’s no feedback loop to understand if they’re working or if something needs to change.
Mistake #3: Pretty Delivered Decks, No Actionable Insights
Those well-designed persona deliverables look great, but when they aren’t tied to briefs, citations, trust signals, your content calendar, etc., they end up siloed from production. If a persona can’t shape a prompt or a page, it won’t shape any of your outcomes.
In addition to the fact classic personas weren’t built to implement across your search strategy, AI has shifted us from optimizing for intent to optimizing for identity and trust. In last week’s memo I shared the following:
The most significant, stand-out finding from that study: People use AI Overviews to get oriented and save time. Then, for any search that involves a transaction or high-stakes decision-making, searchers validate outside Google, usually with trusted brands or authority domains.
Old world of search optimization: Queries signaled intent. You ranked a page that matched the keyword and intent behind it, and your brand would catch the click. Personas were optional.
New world of search optimization: Prompts expose people, and AI changes how we search. Marketers aren’t just optimizing for search intent or demographics; we’re also optimizing for behavior.
Long AI prompts don’t just say what the user intends – they often reveal who is asking and what constraints or background of knowledge they bring.
For example, if a user prompts ChatGPT something like “I’m a healthcare compliance officer at a mid-sized hospital. Can you draft a checklist for evaluating new SaaS vendors, making sure it covers HIPAA regulations and costs under $50K a year,” then ChatGPT would have background information about the user’s general compliance needs, budget ceilings, risk tolerance, and preferred content formats.
AI systems then personalize summaries and citations around that context.
If your content doesn’t meet the persona’s trust requirements or output preference, it won’t be surfaced.
What that means in practice:
- Prompts → identity signals. “As a solo marketer on a $2,000 budget…” or “for EU users under GDPR…” = role, constraints, and risk baked into the query.
- Trust beats length. Classic search results are clicked on, but only when pages show the trust scaffolding a given persona needs for a specific query.
- Format matters. Some personas want TL;DR and tables; others need demos, community validation (YouTube/Reddit), or primary sources.
So, here’s what to do about it.
You don’t need a five or six-figure agency study (although those are nice to have).
You need:
- A collection of your already-existing data.
- A repeatable process, not a static file.
- A way to tie personas directly into briefs and prompts.
Turning your own existing data into usable user personas for SEO will equip you to tie personas directly to content briefs and SEO workflows.
Before you start collecting this data, set up an organized way to store it: Google Sheets, Notion, Airtable – whatever your team prefers. Store your custom persona prompt cards there, too, and you can copy and paste from there into ChatGPT & Co. as needed.
The work below isn’t for the faint of heart, but it will change how you prompt LLMs in your AI-powered workflows and your SEO-focused webpages for the better.
- Collect and cluster data.
- Draft persona prompt cards.
- Calibrate in ChatGPT & Co.
- Validate with real-world signals.
You’re going to mine several data sources that you already have, both qualitative and quantitative.
Keep in mind, being sloppy during this step means you will not have a good base for an “LLM ready” persona prompt card, which I’ll discuss in Step 2.
Attributes to capture for an “LLM-ready persona”:
- Jobs-to-be-done (top 3).
- Role and seniority.
- Buying triggers + blockers (think budget, IT/legal constraints, risk).
- 10-20 example questions at TOFU, MOFU, BOFU stages.
- Trust cues (creators, domains, formats).
- Output preferences (depth, format, tone).
Where AIO validation style data comes in:
Last week, we discussed four distinct AIO intent validations verified within the AIO usability study: Efficiency-first/Trust-driven/Comparative/Skeptical rejection.
If you want to incorporate this in your persona research – and I’d advise that you should – you’re going to look for:
- Hesitation triggers across interactions with your brand: What makes them pause or refine their question (whether on a sales call or a heat map recording).
- Click-out anchors: Which authority brands they use to validate (PayPal, NIH, Mayo Clinic, Stripe, KBB, etc.); use Sparktoro to find this information.
- Evidence threshold: What proof ends hesitation for your user or different personas? (Citations, official terminology, dated reviews, side-by-side tables, videos).
- Device/age nuance: Younger and mobile users → faster AIO acceptance; older cohorts → blue links and authority domains win clicks.
Below, I’ll walk you through where to find this information.
Qualitative Inputs
1. Your GSC queries hold a wealth of info. Split by TOFU/MOFU/BOFU, branded vs non-branded, and country. Then, use a regex to map question-style queries and see who’s really searching at each stage.
Below is the regex I like to use, which I discussed in Is AI cutting into your SEO conversions?. It also works for this task:
(?i)^(who|what|why|how|when|where|which|can|does|is|are|should|guide|tutorial|course|learn|examples?|definition|meaning|checklist|framework|template|tips?|ideas?|best|top|list(?:s)?|comparison|vs|difference|benefits|advantages|alternatives)b.*
2. On-Site Search Logs. These are the records of what visitors type into your website’s own search bar (not Google).
Extract exact phrasing of problems and “missing content” signals (like zero results, refined searches, or high exits/no clicks).
Plus, the wording visitors use reveals jobs-to-be-done, constraints, and vocabulary you should mirror on the page. Flag repeat questions as latent questions to resolve.
3. Support Tickets, CRM Notes, Win/Loss Analysis. Convert objections, blockers, and “how do I…” threads into searchable intents and hesitation themes.
Mine the following data from your records:
- Support: Ticket titles, first message, last agent note, resolution summary.
- CRM: Opportunity notes, metrics, decision criteria, lost-reason text.
- Win/Loss: Objection snapshots, competitor cited, decision drivers, de-risking asks.
- Context (if available): buyer role, segment (SMB/MM/ENT), region, product line, funnel stage.
Once gathered, compile and analyze to distill patterns.
Qualitative Inputs
1. Your sales calls and customer success notes are a wealth of information.
Use AI to analyze transcripts and/or notes to highlight jobs-to-be-done, triggers, blockers, and decision criteria in your customer’s own words.
2. Reddit and social media discussions.
This is where your buyers actually compare options and validate claims; capture the authority anchors (brands/domains) they trust.
3. Community/Slack spaces, email newsletter replies, article comments, short post-purchase or signup surveys.
Mine recurring “stuck points” and vocabulary you should mirror. Bucket recurring themes together and correlate across other data.
Pro tip: Use your topic map as the semantic backbone for all qualitative synthesis – discussed in depth in how to operationalize topic-first SEO. You’d start by locking the parent topics, then layer your personas as lenses: For each parent topic, fan out subtopics by persona, funnel stage, and the “people × problems” you pull from sales calls, CS notes, Reddit/LinkedIn, and community threads. Flag zero-volume/fringe questions on your map as priorities; they deepen authority and often resolve the hesitation themes your notes reveal.
After clustering pain points and recurring queries, you can take it one step further to tag each cluster with an AIO pattern by looking for:
- Short dwell + 0–1 scroll + no refinements → Efficiency-first validations.
- Longer dwell + multiple scrolls + hesitation language + authority click-outs → Trust-driven validations.
- Four to five scrolls + multiple tabs (YouTube/Reddit/vendor) → Comparative validations.
- Minimal AIO engagement + direct authority clicks (gov/medical/finance) → Skeptical rejection.
Not every team can run a full-blown usability study of the search results for targeted queries and topics, but you can infer many of these behavioral patterns through heatmaps of your own pages that have strong organic visibility.
2. Draft Persona Prompt Cards
Next up, you’ll take this data to inform creating a persona card.
A persona card is a one-page, ready-to-go snapshot of a target user segment that your marketing/SEO team can act on.
Unlike empty or demographic-heavy personas, a persona card ties jobs-to-be-done, constraints, questions, and trust cues directly to how you brief pages, structure proofs, and prompt LLMs.
A persona card ensures your pages and prompts match identity + trust requirements.
What you’re going to do in this step is convert each data-based persona cluster into a one-pager designed to be embedded directly into LLM prompts.
Include input patterns you expect from that persona – and the output format they’d likely want.
Optimizing Prompt Selection for Target Audience Engagement
Reusable Template: Persona Prompt Card
Drop this at the top of a ChatGPT conversation or save as a snippet.
This is an example template below based on the Growth Memo audience specifically, so you’ll need to not only modify it for your needs, but also tweak it per persona.
You are Kevin Indig advising a [ROLE, SENIORITY] en un [COMPANY TYPE, SIZE, LOCATION]. Objetivo: [Top 1–2 goals tied to KPIs and timeline]Contexto: [Market, constraints, budget guardrails, compliance/IT notes]Estilo de pregunta de la persona: [Example inputs they’d type; tone & jargon tolerance] Formato de respuesta: - Comience con un TL de 3 batales; DR. - Luego, dé un libro de jugadas con 5-7 pasos. - Incluya 2 puntos de prueba (puntos de referencia/estudios de caso) y 1 calculadora/plantilla. - Riesgos de bandera y compensaciones explícitamente. - mantener [brevity/depth]; [bullets/narrative]; incluir [table/chart] Si es útil. Qué evitar: [Banned claims, fluff, vendor speak] Citas: prefiere [domains/creators] e investigación original cuando sea posible.
Ejemplo de conjuntos de atributos utilizando la audiencia de memorando de crecimiento
Use esta tarjeta como punto de partida, luego llénela con sus datos.
A continuación se muestra un ejemplo de la tarjeta de inmediato con atributos llenos para uno de los perfiles de clientes ideales (ICP) para la audiencia de memo de crecimiento.
You are Kevin Indig advising an SEO Lead (Senior) at a Mid-Market B2B SaaS (US/EU). Objective: Protect and grow organic pipeline in the AI-search era; drive qualified trials/demos in Q4; build durable topic authority. Context: Competitive category; CMS constraints + limited Eng bandwidth; GDPR/CCPA; security/legal review for pages; budget ≤ $8,000/mo for content + tools; stakeholders: VP Marketing, Content Lead, PMM, RevOps. Persona question style: “How do I measure topic performance vs keywords?”, “How do I structure entity-based internal linking?”, “What KPIs prove AIO exposure matters?”, “Regex for TOFU/MOFU/BOFU?”, “How to brief comparison pages that AIO cites?” Tone: precise, low-fluff, technical. AIO validation profile: - Dominant pattern(s): Trust-driven (primary), Comparative (frameworks/tools); Skeptical for YMYL claims. - Hesitation triggers: Black-box vendor claims; non-replicable methods; missing citations; unclear risk/effort. - Click-out anchors: Google Search Central & docs, schema.org, reputable research (Semrush/Ahrefs/SISTRIX/seoClarity), Pew/Ofcom, credible case studies, engineering/product docs. - SERP feature bias: Skims AIO/snippets to frame, validates via organic authority + primary sources; uses YouTube for demos; largely ignores Ads. - Evidence threshold: Methodology notes, datasets/replication steps, benchmarks, decision tables, risk trade-offs. Answer format: - Start with a three-bullet TL;DR. - Then give a numbered playbook with 5-7 steps. - Include 2 proof points (benchmarks/case studies) and 1 calculator/template. - Flag risks and trade-offs explicitly. - Keep to brevity + bullets; include a table/chart if useful. Proof kit to include on-page: Methodology & data provenance; decision table (framework/tool choice); “best for / not for”; internal-linking map or schema snippet; last-reviewed date; citations to Google docs/primary research; short demo or worksheet (e.g., Topic Coverage Score or KPI tree). What to avoid: Vendor-speak; outdated screenshots; cherry-picked wins; unverifiable stats; hand-wavy “AI magic.” Citations: Prefer Google Search Central/docs, schema.org, original studies/datasets; reputable tool research (Semrush, Ahrefs, SISTRIX, seoClarity); peer case studies with numbers. Success signals to watch: Topic-level lift (impressions/CTR/coverage), assisted conversions from topic clusters, AIO/snippet presence for key topics, authority referrals, demo starts from comparison hubs, reduced content decay, improved crawl/indexation on priority clusters.
Su objetivo aquí es demostrar que las tarjetas rápidas de la persona realmente producen respuestas útiles, y aprender qué evidencia necesita cada persona.
Cree un perfil de instrucciones personalizado por persona, o guarde cada tarjeta de solicitud de persona como un fragmento de inmediato que puede prepender.
Corre 10-15 consultas reales por persona. Marque respuestas sobre claridad, escaneabilidad, credibilidad y diferenciación a su estándar.
Cómo ejecutar la calibración de la tarjeta rápida:
- Configuración: Guarde una tarjeta rápida por persona.
- Establecimiento de evaluación: 10-15 real Consultas/Personidades en etapas de tofu/mofu/bofu, incluidas dos o tres consultas basadas en el cumplimiento o basadas en el cumplimiento, tres a cuatro comparaciones y tres o cuatro How-Tos rápidos.
- Solicitar estructura: Requiere TL; DR → Libro de jugadas numerado → Tabla → Riesgos → Citas (según la tarjeta).
- Modifíquelo: Agregar restricciones y variantes de ubicación; Pregunte a la misma consulta dos formas de probar la consistencia.
Una vez que ejecute consultas de muestra para verificar la claridad y la credibilidad, modifique o actualice su tarjeta de personal según sea necesario: agregue los anclajes de confianza faltantes o evidencia que el modelo necesitaba.
Guarde los resultados ganadores como formas de guiar sus informes que puede pegar en borradores.
Log recurrente Misses (estadísticas alucinadas, reclamos sin fecha) como verificaciones de aceptación para la producción.
Luego, haga esto para otros LLM que usa su audiencia. Por ejemplo, si su audiencia se inclina fuertemente hacia el uso de perplejidad. Asegúrese de ejecutar también las salidas de tarjeta de inmediato en el modo AI de Google.
Mire las tendencias de búsqueda de marca, las conversiones asistidas y las referencias que no son de Google para ver si la influencia aparece donde se esperaba cuando publica activos sintonizados por personal.
Y asegúrese de medir la elevación por tema, no solo por página: rendimiento del segmento por clúster de temas (gsc regex o ga4 dimensión de tema). La operación de su estrategia de SEO de su tema discute cómo hacer esto.
Tenga lo siguiente en mente al revisar las señales del mundo real:
- Revise al 30/60/90 días después de la nave y por el clúster de temas.
- Si las páginas impulsadas por la confianza muestran altas conversiones de desplazamiento/baja → Citas de agregar/actualizar y revisiones y citas de expertos.
- Si las páginas comparativas obtienen registros de demostraciones de productos / ventas bajas → Agregar video de demostración corto, secciones «Mejor para / no para» y CTA más claros.
- Si las páginas de eficiencia primero se pierden en AIO/Fragmentos → Apriete TL; DR, simplifique las tablas, agregue el esquema.
- Si las páginas con rechazo escéptico producen el tráfico de la autoridad pero no hay elevación → Considere buscar asociaciones de autoridad.
- Lo más importante: rehace el ejercicio cada 60-90 días y coincida con su nuevo con personas viejas para iterar hacia el ideal.
La creación de personajes de los usuarios para SEO vale la pena, y puede ser factible y rápido mediante el uso de datos internos y soporte de LLM.
Te desafío a que comiences con una personalidad Lean esta semana para probar este enfoque. Refina y amplíe su enfoque en función de los resultados que ve.
Pero si planea tomar este proyecto de construcción de personas, evite estos pasos en falso comunes:
- Creación de PDF ordenados con cero beneficios a largo plazo: Personas que no especifican intentos de búsqueda básicos, puntos débiles y patrones de intención de AIO no moverán el comportamiento.
- Ganar cada característica de SERP: Esta es una pérdida de tiempo. Optimice su contenido para la superficie adecuada para los patrones de comportamiento dominantes de sus usuarios objetivo.
- Iglorando la vacilación: La duda es tu mayor señal. Si no lo resuelve en la página, el clic muere en otro lugar.
- Demografía sobre trabajos a ser hechos: Centrarse en las características de la identidad sin incorporar patrones de comportamiento es la antigua forma.
Imagen Feaded: Paulo Bobita/Search Engine Journal
